├── README.md ├── code └── demo.m ├── data └── imgs │ ├── 0_0_272.jpg │ └── 0_0_775.jpg ├── deeplab-caffe ├── .Doxyfile ├── .gitignore ├── .travis.yml ├── CAFFE_README.md ├── CMakeLists.txt ├── CONTRIBUTING.md ├── CONTRIBUTORS.md ├── INSTALL.md ├── LICENSE ├── Makefile ├── Makefile.config.example ├── README.md ├── caffe.cloc ├── cmake │ ├── ConfigGen.cmake │ ├── Cuda.cmake │ ├── Dependencies.cmake │ ├── External │ │ ├── gflags.cmake │ │ └── glog.cmake │ ├── Misc.cmake │ ├── Modules │ │ ├── FindAtlas.cmake │ │ ├── FindGFlags.cmake │ │ ├── FindGlog.cmake │ │ ├── FindLAPACK.cmake │ │ ├── FindLMDB.cmake │ │ ├── FindLevelDB.cmake │ │ ├── FindMKL.cmake │ │ ├── FindMatlabMex.cmake │ │ ├── FindNumPy.cmake │ │ ├── FindOpenBLAS.cmake │ │ ├── FindSnappy.cmake │ │ └── FindvecLib.cmake │ ├── ProtoBuf.cmake │ ├── Summary.cmake │ ├── Targets.cmake │ ├── Templates │ │ ├── CaffeConfig.cmake.in │ │ ├── CaffeConfigVersion.cmake.in │ │ └── caffe_config.h.in │ ├── Utils.cmake │ └── lint.cmake ├── copyweights.m ├── densecrf │ ├── Makefile │ ├── README.md │ ├── libDenseCRF │ │ ├── bipartitedensecrf.cpp │ │ ├── densecrf.cpp │ │ ├── densecrf.h │ │ ├── fastmath.h │ │ ├── filter.cpp │ │ ├── permutohedral.cpp │ │ ├── permutohedral.h │ │ ├── sse_defs.h │ │ ├── util.cpp │ │ └── util.h │ ├── my_script │ │ ├── DownSampleFeature.m │ │ ├── GetDenseCRFResult.m │ │ ├── GetDenseCRFResult_CrossValidate.m │ │ ├── GetDenseCRFResult_CrossValidate_erodeGT.m │ │ ├── GetDenseCRFResult_erodeGT.m │ │ ├── LoadBinFile.m │ │ ├── My_GetDenseCRFResult.m │ │ ├── SaveBinFile.m │ │ ├── SaveJpgToBin.m │ │ ├── SaveJpgToPPM.m │ │ ├── SaveMatAsBin.m │ │ ├── SavePngToPPM.m │ │ └── pascal_seg_colormap.mat │ ├── prog_refine_pascal │ ├── prog_refine_pascal_v4 │ ├── prog_test_densecrf │ ├── refine_pascal │ │ ├── dense_inference.cpp │ │ └── dense_inference.h │ ├── refine_pascal_v4 │ │ └── dense_inference.cpp │ ├── test_densecrf │ │ └── simple_dense_inference.cpp │ └── util │ │ └── Timer.h ├── densecrf_float │ ├── Makefile │ ├── README.md │ ├── libDenseCRF │ │ ├── bipartitedensecrf.cpp │ │ ├── densecrf.cpp │ │ ├── densecrf.h │ │ ├── fastmath.h │ │ ├── filter.cpp │ │ ├── permutohedral.cpp │ │ ├── permutohedral.h │ │ ├── sse_defs.h │ │ ├── util.cpp │ │ └── util.h │ ├── my_script │ │ ├── DownSampleFeature.m │ │ ├── GetDenseCRFResult.m │ │ ├── GetDenseCRFResult_CrossValidate.m │ │ ├── GetDenseCRFResult_CrossValidate_erodeGT.m │ │ ├── GetDenseCRFResult_erodeGT.m │ │ ├── LoadBinFile.m │ │ ├── SaveBinFile.m │ │ ├── SaveJpgToBin.m │ │ ├── SaveJpgToPPM.m │ │ ├── SaveMatAsBin.m │ │ ├── bin2segimg.m │ │ └── pascal_seg_colormap.mat │ ├── prog_refine_pascal │ ├── prog_refine_pascal_v4 │ ├── prog_test_densecrf │ ├── refine_pascal │ │ ├── dense_inference.cpp │ │ └── dense_inference.h │ ├── refine_pascal_v4 │ │ └── dense_inference.cpp │ ├── test_densecrf │ │ └── simple_dense_inference.cpp │ └── util │ │ └── Timer.h ├── docs │ ├── CMakeLists.txt │ ├── CNAME │ ├── README.md │ ├── _config.yml │ ├── _layouts │ │ └── default.html │ ├── development.md │ ├── images │ │ ├── GitHub-Mark-64px.png │ │ └── caffeine-icon.png │ ├── index.md │ ├── install_apt.md │ ├── install_osx.md │ ├── install_yum.md │ ├── installation.md │ ├── model_zoo.md │ ├── multigpu.md │ ├── performance_hardware.md │ ├── stylesheets │ │ ├── pygment_trac.css │ │ ├── reset.css │ │ └── styles.css │ └── tutorial │ │ ├── convolution.md │ │ ├── data.md │ │ ├── fig │ │ ├── .gitignore │ │ ├── backward.jpg │ │ ├── forward.jpg │ │ ├── forward_backward.png │ │ ├── layer.jpg │ │ └── logreg.jpg │ │ ├── forward_backward.md │ │ ├── index.md │ │ ├── interfaces.md │ │ ├── layers.md │ │ ├── loss.md │ │ ├── net_layer_blob.md │ │ └── solver.md ├── examples │ ├── 00-classification.ipynb │ ├── 01-learning-lenet.ipynb │ ├── 02-fine-tuning.ipynb │ ├── CMakeLists.txt │ ├── brewing-logreg.ipynb │ ├── cifar10 │ │ ├── cifar10_full.prototxt │ │ ├── cifar10_full_sigmoid_solver.prototxt │ │ ├── cifar10_full_sigmoid_solver_bn.prototxt │ │ ├── cifar10_full_sigmoid_train_test.prototxt │ │ ├── cifar10_full_sigmoid_train_test_bn.prototxt │ │ ├── cifar10_full_solver.prototxt │ │ ├── cifar10_full_solver_lr1.prototxt │ │ ├── cifar10_full_solver_lr2.prototxt │ │ ├── cifar10_full_train_test.prototxt │ │ ├── cifar10_quick.prototxt │ │ ├── cifar10_quick_solver.prototxt │ │ ├── cifar10_quick_solver_lr1.prototxt │ │ ├── cifar10_quick_train_test.prototxt │ │ ├── convert_cifar_data.cpp │ │ ├── create_cifar10.sh │ │ ├── readme.md │ │ ├── train_full.sh │ │ ├── train_full_sigmoid.sh │ │ ├── train_full_sigmoid_bn.sh │ │ └── train_quick.sh │ ├── cpp_classification │ │ ├── classification.cpp │ │ └── readme.md │ ├── detection.ipynb │ ├── feature_extraction │ │ ├── imagenet_val.prototxt │ │ └── readme.md │ ├── finetune_flickr_style │ │ ├── assemble_data.py │ │ ├── flickr_style.csv.gz │ │ ├── readme.md │ │ └── style_names.txt │ ├── finetune_pascal_detection │ │ ├── pascal_finetune_solver.prototxt │ │ └── pascal_finetune_trainval_test.prototxt │ ├── hdf5_classification │ │ ├── nonlinear_auto_test.prototxt │ │ ├── nonlinear_auto_train.prototxt │ │ ├── nonlinear_train_val.prototxt │ │ └── train_val.prototxt │ ├── imagenet │ │ ├── create_imagenet.sh │ │ ├── make_imagenet_mean.sh │ │ ├── readme.md │ │ ├── resume_training.sh │ │ └── train_caffenet.sh │ ├── images │ │ ├── cat.jpg │ │ ├── cat_gray.jpg │ │ └── fish-bike.jpg │ ├── mnist │ │ ├── convert_mnist_data.cpp │ │ ├── create_mnist.sh │ │ ├── lenet.prototxt │ │ ├── lenet_adadelta_solver.prototxt │ │ ├── lenet_auto_solver.prototxt │ │ ├── lenet_consolidated_solver.prototxt │ │ ├── lenet_multistep_solver.prototxt │ │ ├── lenet_solver.prototxt │ │ ├── lenet_solver_adam.prototxt │ │ ├── lenet_solver_rmsprop.prototxt │ │ ├── lenet_train_test.prototxt │ │ ├── mnist_autoencoder.prototxt │ │ ├── mnist_autoencoder_solver.prototxt │ │ ├── mnist_autoencoder_solver_adadelta.prototxt │ │ ├── mnist_autoencoder_solver_adagrad.prototxt │ │ ├── mnist_autoencoder_solver_nesterov.prototxt │ │ ├── readme.md │ │ ├── train_lenet.sh │ │ ├── train_lenet_adam.sh │ │ ├── train_lenet_consolidated.sh │ │ ├── train_lenet_rmsprop.sh │ │ ├── train_mnist_autoencoder.sh │ │ ├── train_mnist_autoencoder_adadelta.sh │ │ ├── train_mnist_autoencoder_adagrad.sh │ │ └── train_mnist_autoencoder_nesterov.sh │ ├── net_surgery.ipynb │ ├── net_surgery │ │ ├── bvlc_caffenet_full_conv.prototxt │ │ └── conv.prototxt │ ├── pycaffe │ │ ├── caffenet.py │ │ ├── layers │ │ │ └── pyloss.py │ │ └── linreg.prototxt │ ├── siamese │ │ ├── convert_mnist_siamese_data.cpp │ │ ├── create_mnist_siamese.sh │ │ ├── mnist_siamese.ipynb │ │ ├── mnist_siamese.prototxt │ │ ├── mnist_siamese_solver.prototxt │ │ ├── mnist_siamese_train_test.prototxt │ │ ├── readme.md │ │ └── train_mnist_siamese.sh │ └── web_demo │ │ ├── app.py │ │ ├── exifutil.py │ │ ├── readme.md │ │ ├── requirements.txt │ │ └── templates │ │ └── index.html ├── include │ └── caffe │ │ ├── blob.hpp │ │ ├── caffe.hpp │ │ ├── common.cuh │ │ ├── common.hpp │ │ ├── data_reader.hpp │ │ ├── data_transformer.hpp │ │ ├── filler.hpp │ │ ├── internal_thread.hpp │ │ ├── layer.hpp │ │ ├── layer_factory.hpp │ │ ├── layers │ │ ├── absval_layer.hpp │ │ ├── accuracy_layer.hpp │ │ ├── adaptive_bias_channel_layer.hpp │ │ ├── argmax_layer.hpp │ │ ├── base_conv_layer.hpp │ │ ├── base_data_layer.hpp │ │ ├── batch_norm_layer.hpp │ │ ├── batch_reindex_layer.hpp │ │ ├── bias_channel_layer.hpp │ │ ├── bias_layer.hpp │ │ ├── bnll_layer.hpp │ │ ├── concat_layer.hpp │ │ ├── contour_accuracy_layer.hpp │ │ ├── contrastive_loss_layer.hpp │ │ ├── conv_layer.hpp │ │ ├── cudnn_conv_layer.hpp │ │ ├── cudnn_lcn_layer.hpp │ │ ├── cudnn_lrn_layer.hpp │ │ ├── cudnn_pooling_layer.hpp │ │ ├── cudnn_relu_layer.hpp │ │ ├── cudnn_sigmoid_layer.hpp │ │ ├── cudnn_softmax_layer.hpp │ │ ├── cudnn_tanh_layer.hpp │ │ ├── data_layer.hpp │ │ ├── deconv_layer.hpp │ │ ├── densecrf_layer.hpp │ │ ├── domain_transform_forward_only_layer.hpp │ │ ├── domain_transform_layer.hpp │ │ ├── dropout_layer.hpp │ │ ├── dummy_data_layer.hpp │ │ ├── eltwise_layer.hpp │ │ ├── elu_layer.hpp │ │ ├── embed_layer.hpp │ │ ├── euclidean_loss_layer.hpp │ │ ├── exp_layer.hpp │ │ ├── filter_layer.hpp │ │ ├── flatten_layer.hpp │ │ ├── get_data_dim_layer.hpp │ │ ├── hdf5_data_layer.hpp │ │ ├── hdf5_output_layer.hpp │ │ ├── hinge_loss_layer.hpp │ │ ├── im2col_layer.hpp │ │ ├── image_cls_data_layer.hpp │ │ ├── image_data_layer.hpp │ │ ├── image_salobj_data_layer.hpp │ │ ├── image_seg_data_layer.hpp │ │ ├── infogain_loss_layer.hpp │ │ ├── inner_product_layer.hpp │ │ ├── interp_layer.hpp │ │ ├── log_layer.hpp │ │ ├── loss_layer.hpp │ │ ├── lrn_layer.hpp │ │ ├── mat_read_layer.hpp │ │ ├── mat_write_layer.hpp │ │ ├── memory_data_layer.hpp │ │ ├── multinomial_logistic_loss_layer.hpp │ │ ├── mvn_layer.hpp │ │ ├── neuron_layer.hpp │ │ ├── pooling_layer.hpp │ │ ├── power_layer.hpp │ │ ├── prelu_layer.hpp │ │ ├── python_layer.hpp │ │ ├── reduction_layer.hpp │ │ ├── relu_layer.hpp │ │ ├── reshape_layer.hpp │ │ ├── scale_layer.hpp │ │ ├── seg_accuracy_layer.hpp │ │ ├── sigmoid_cross_entropy_loss_layer.hpp │ │ ├── sigmoid_layer.hpp │ │ ├── silence_layer.hpp │ │ ├── slice_layer.hpp │ │ ├── softmax_layer.hpp │ │ ├── softmax_loss_layer.hpp │ │ ├── spatial_product_layer.hpp │ │ ├── split_layer.hpp │ │ ├── spp_layer.hpp │ │ ├── tanh_layer.hpp │ │ ├── threshold_layer.hpp │ │ ├── tile_layer.hpp │ │ ├── unique_label_layer.hpp │ │ ├── unpooling_layer.hpp │ │ ├── visual_saliency_layer.hpp │ │ └── window_data_layer.hpp │ │ ├── net.hpp │ │ ├── parallel.hpp │ │ ├── sgd_solvers.hpp │ │ ├── solver.hpp │ │ ├── solver_factory.hpp │ │ ├── syncedmem.hpp │ │ ├── test │ │ ├── test_caffe_main.hpp │ │ └── test_gradient_check_util.hpp │ │ └── util │ │ ├── benchmark.hpp │ │ ├── blocking_queue.hpp │ │ ├── confusion_matrix.hpp │ │ ├── cudnn.hpp │ │ ├── db.hpp │ │ ├── db_leveldb.hpp │ │ ├── db_lmdb.hpp │ │ ├── densecrf_pairwise.hpp │ │ ├── densecrf_util.hpp │ │ ├── device_alternate.hpp │ │ ├── format.hpp │ │ ├── gpu_util.cuh │ │ ├── hdf5.hpp │ │ ├── im2col.hpp │ │ ├── insert_splits.hpp │ │ ├── interp.hpp │ │ ├── io.hpp │ │ ├── math_functions.hpp │ │ ├── matio_io.hpp │ │ ├── mkl_alternate.hpp │ │ ├── permutohedral.hpp │ │ ├── rng.hpp │ │ ├── signal_handler.h │ │ └── upgrade_proto.hpp ├── matlab │ ├── +caffe │ │ ├── +test │ │ │ ├── test_io.m │ │ │ ├── test_net.m │ │ │ └── test_solver.m │ │ ├── Blob.m │ │ ├── Layer.m │ │ ├── Net.m │ │ ├── Solver.m │ │ ├── get_net.m │ │ ├── get_solver.m │ │ ├── imagenet │ │ │ └── ilsvrc_2012_mean.mat │ │ ├── io.m │ │ ├── private │ │ │ ├── CHECK.m │ │ │ ├── CHECK_FILE_EXIST.m │ │ │ ├── caffe_.cpp │ │ │ └── is_valid_handle.m │ │ ├── reset_all.m │ │ ├── run_tests.m │ │ ├── set_device.m │ │ ├── set_mode_cpu.m │ │ ├── set_mode_gpu.m │ │ └── version.m │ ├── CMakeLists.txt │ ├── demo │ │ └── classification_demo.m │ ├── hdf5creation │ │ ├── .gitignore │ │ ├── demo.m │ │ └── store2hdf5.m │ ├── my_script │ │ ├── 0_0_147_blob_0.mat │ │ ├── 0_0_272.png │ │ ├── 1.png │ │ ├── 2007_000032_blob_0.mat │ │ ├── 2007_000762.png │ │ ├── 75.png │ │ ├── 9990.png │ │ ├── AppendPrefixPostfix.m │ │ ├── DownSampleGtBboxErodeBin.m │ │ ├── EvalCRF_Boundary.m │ │ ├── EvalSaliencyResults.m │ │ ├── EvalSegResults.m │ │ ├── EvalSegResults_CrossValidate.m │ │ ├── EvalSegResults_CrossValidate_erodeGT.m │ │ ├── GetCocoCategories.m │ │ ├── GetImglistForCaffe.m │ │ ├── GetList.m │ │ ├── GetMSRAopts.m │ │ ├── GetPASCALContourDataopts.m │ │ ├── GetSaliencyOpts.m │ │ ├── GetSegmentationResult.m │ │ ├── GetVOCopts.m │ │ ├── LoadBinFile.m │ │ ├── MyMSRAevalseg.m │ │ ├── MySaliencyEval.m │ │ ├── MyVOCevalseg.m │ │ ├── MyVOCevalsegBoundary.m │ │ ├── PasteSoftSegmentBboxByArea.m │ │ ├── PlotBoundaryAccuracy.m │ │ ├── SaveBinFile.m │ │ ├── SavePngAsRawPng.m │ │ ├── SetupEnv.m │ │ ├── ShowGT.m │ │ ├── ShowResults.m │ │ ├── SortSegmentationByBbox.m │ │ ├── TransformBerkeleyVOC2011Annot.m │ │ ├── TransformBerkeleyVOC2011Annot_InstanceLevel_imerode_saveAsBin.m │ │ ├── TransformSegToProb.m │ │ ├── convertMatToPNG.m │ │ ├── cross_avgIOU_voc12_features_fc8DownSample118.txtcross_avgIOU_al_d_ │ │ ├── funcs │ │ │ ├── CalFmeasure.m │ │ │ ├── CalMAE.m │ │ │ ├── CalMeanMAE.m │ │ │ ├── CalMeanPR.m │ │ │ ├── CalPR.m │ │ │ └── DrawPRCurve.m │ │ ├── pascal_seg_colormap.mat │ │ ├── result.png │ │ └── saliency_colormap.mat │ └── resnet101_net_surgery.m ├── python │ ├── CMakeLists.txt │ ├── caffe │ │ ├── __init__.py │ │ ├── _caffe.cpp │ │ ├── classifier.py │ │ ├── detector.py │ │ ├── draw.py │ │ ├── imagenet │ │ │ └── ilsvrc_2012_mean.npy │ │ ├── io.py │ │ ├── net_spec.py │ │ ├── pycaffe.py │ │ └── test │ │ │ ├── test_io.py │ │ │ ├── test_layer_type_list.py │ │ │ ├── test_net.py │ │ │ ├── test_net_spec.py │ │ │ ├── test_python_layer.py │ │ │ ├── test_python_layer_with_param_str.py │ │ │ └── test_solver.py │ ├── classify.py │ ├── detect.py │ ├── draw_net.py │ └── requirements.txt ├── scripts │ ├── build_docs.sh │ ├── copy_notebook.py │ ├── cpp_lint.py │ ├── deploy_docs.sh │ ├── download_model_binary.py │ ├── download_model_from_gist.sh │ ├── gather_examples.sh │ ├── travis │ │ ├── travis_build_and_test.sh │ │ ├── travis_install.sh │ │ └── travis_setup_makefile_config.sh │ └── upload_model_to_gist.sh ├── src │ ├── caffe │ │ ├── CMakeLists.txt │ │ ├── blob.cpp │ │ ├── common.cpp │ │ ├── data_reader.cpp │ │ ├── data_transformer.cpp │ │ ├── internal_thread.cpp │ │ ├── layer.cpp │ │ ├── layer_factory.cpp │ │ ├── layers │ │ │ ├── absval_layer.cpp │ │ │ ├── absval_layer.cu │ │ │ ├── accuracy_layer.cpp │ │ │ ├── adaptive_bias_channel_layer.cpp │ │ │ ├── argmax_layer.cpp │ │ │ ├── base_conv_layer.cpp │ │ │ ├── base_data_layer.cpp │ │ │ ├── base_data_layer.cu │ │ │ ├── batch_norm_layer.cpp │ │ │ ├── batch_norm_layer.cu │ │ │ ├── batch_reindex_layer.cpp │ │ │ ├── batch_reindex_layer.cu │ │ │ ├── bias_channel_layer.cpp │ │ │ ├── bias_layer.cpp │ │ │ ├── bias_layer.cu │ │ │ ├── bnll_layer.cpp │ │ │ ├── bnll_layer.cu │ │ │ ├── concat_layer.cpp │ │ │ ├── concat_layer.cu │ │ │ ├── contour_accuracy_layer.cpp │ │ │ ├── contrastive_loss_layer.cpp │ │ │ ├── contrastive_loss_layer.cu │ │ │ ├── conv_layer.cpp │ │ │ ├── conv_layer.cu │ │ │ ├── cudnn_conv_layer.cpp │ │ │ ├── cudnn_conv_layer.cu │ │ │ ├── cudnn_lcn_layer.cpp │ │ │ ├── cudnn_lcn_layer.cu │ │ │ ├── cudnn_lrn_layer.cpp │ │ │ ├── cudnn_lrn_layer.cu │ │ │ ├── cudnn_pooling_layer.cpp │ │ │ ├── cudnn_pooling_layer.cu │ │ │ ├── cudnn_relu_layer.cpp │ │ │ ├── cudnn_relu_layer.cu │ │ │ ├── cudnn_sigmoid_layer.cpp │ │ │ ├── cudnn_sigmoid_layer.cu │ │ │ ├── cudnn_softmax_layer.cpp │ │ │ ├── cudnn_softmax_layer.cu │ │ │ ├── cudnn_tanh_layer.cpp │ │ │ ├── cudnn_tanh_layer.cu │ │ │ ├── data_layer.cpp │ │ │ ├── deconv_layer.cpp │ │ │ ├── deconv_layer.cu │ │ │ ├── densecrf_layer.cpp │ │ │ ├── domain_transform_forward_only_layer.cpp │ │ │ ├── domain_transform_forward_only_layer.cu │ │ │ ├── domain_transform_layer.cpp │ │ │ ├── domain_transform_layer.cu │ │ │ ├── dropout_layer.cpp │ │ │ ├── dropout_layer.cu │ │ │ ├── dummy_data_layer.cpp │ │ │ ├── eltwise_layer.cpp │ │ │ ├── eltwise_layer.cu │ │ │ ├── elu_layer.cpp │ │ │ ├── elu_layer.cu │ │ │ ├── embed_layer.cpp │ │ │ ├── embed_layer.cu │ │ │ ├── euclidean_loss_layer.cpp │ │ │ ├── euclidean_loss_layer.cu │ │ │ ├── exp_layer.cpp │ │ │ ├── exp_layer.cu │ │ │ ├── filter_layer.cpp │ │ │ ├── filter_layer.cu │ │ │ ├── flatten_layer.cpp │ │ │ ├── get_data_dim_layer.cpp │ │ │ ├── hdf5_data_layer.cpp │ │ │ ├── hdf5_data_layer.cu │ │ │ ├── hdf5_output_layer.cpp │ │ │ ├── hdf5_output_layer.cu │ │ │ ├── hinge_loss_layer.cpp │ │ │ ├── im2col_layer.cpp │ │ │ ├── im2col_layer.cu │ │ │ ├── image_cls_data_layer.cpp │ │ │ ├── image_data_layer.cpp │ │ │ ├── image_salobj_data_layer.cpp │ │ │ ├── image_seg_data_layer.cpp │ │ │ ├── infogain_loss_layer.cpp │ │ │ ├── inner_product_layer.cpp │ │ │ ├── inner_product_layer.cu │ │ │ ├── interp_layer.cpp │ │ │ ├── log_layer.cpp │ │ │ ├── log_layer.cu │ │ │ ├── loss_layer.cpp │ │ │ ├── lrn_layer.cpp │ │ │ ├── lrn_layer.cu │ │ │ ├── mat_read_layer.cpp │ │ │ ├── mat_write_layer.cpp │ │ │ ├── memory_data_layer.cpp │ │ │ ├── multinomial_logistic_loss_layer.cpp │ │ │ ├── mvn_layer.cpp │ │ │ ├── mvn_layer.cu │ │ │ ├── neuron_layer.cpp │ │ │ ├── pooling_layer.cpp │ │ │ ├── pooling_layer.cu │ │ │ ├── power_layer.cpp │ │ │ ├── power_layer.cu │ │ │ ├── prelu_layer.cpp │ │ │ ├── prelu_layer.cu │ │ │ ├── reduction_layer.cpp │ │ │ ├── reduction_layer.cu │ │ │ ├── relu_layer.cpp │ │ │ ├── relu_layer.cu │ │ │ ├── reshape_layer.cpp │ │ │ ├── scale_layer.cpp │ │ │ ├── scale_layer.cu │ │ │ ├── seg_accuracy_layer.cpp │ │ │ ├── sigmoid_cross_entropy_loss_layer.cpp │ │ │ ├── sigmoid_layer.cpp │ │ │ ├── sigmoid_layer.cu │ │ │ ├── silence_layer.cpp │ │ │ ├── silence_layer.cu │ │ │ ├── slice_layer.cpp │ │ │ ├── slice_layer.cu │ │ │ ├── softmax_layer.cpp │ │ │ ├── softmax_layer.cu │ │ │ ├── softmax_loss_layer.cpp │ │ │ ├── softmax_loss_layer.cu │ │ │ ├── spatial_product_layer.cpp │ │ │ ├── spatial_product_layer.cu │ │ │ ├── split_layer.cpp │ │ │ ├── split_layer.cu │ │ │ ├── spp_layer.cpp │ │ │ ├── tanh_layer.cpp │ │ │ ├── tanh_layer.cu │ │ │ ├── threshold_layer.cpp │ │ │ ├── threshold_layer.cu │ │ │ ├── tile_layer.cpp │ │ │ ├── tile_layer.cu │ │ │ ├── unique_label_layer.cpp │ │ │ ├── unpooling_layer.cpp │ │ │ ├── unpooling_layer.cu │ │ │ ├── visual_saliency_layer.cpp │ │ │ └── window_data_layer.cpp │ │ ├── net.cpp │ │ ├── parallel.cpp │ │ ├── proto │ │ │ └── caffe.proto │ │ ├── solver.cpp │ │ ├── solvers │ │ │ ├── adadelta_solver.cpp │ │ │ ├── adadelta_solver.cu │ │ │ ├── adagrad_solver.cpp │ │ │ ├── adagrad_solver.cu │ │ │ ├── adam_solver.cpp │ │ │ ├── adam_solver.cu │ │ │ ├── nesterov_solver.cpp │ │ │ ├── nesterov_solver.cu │ │ │ ├── rmsprop_solver.cpp │ │ │ ├── rmsprop_solver.cu │ │ │ ├── sgd_solver.cpp │ │ │ └── sgd_solver.cu │ │ ├── syncedmem.cpp │ │ ├── test │ │ │ ├── CMakeLists.txt │ │ │ ├── test_accuracy_layer.cpp │ │ │ ├── test_argmax_layer.cpp │ │ │ ├── test_batch_norm_layer.cpp │ │ │ ├── test_batch_reindex_layer.cpp │ │ │ ├── test_benchmark.cpp │ │ │ ├── test_bias_channel_layer.cpp │ │ │ ├── test_bias_layer.cpp │ │ │ ├── test_blob.cpp │ │ │ ├── test_caffe_main.cpp │ │ │ ├── test_common.cpp │ │ │ ├── test_concat_layer.cpp │ │ │ ├── test_contrastive_loss_layer.cpp │ │ │ ├── test_convolution_layer.cpp │ │ │ ├── test_data │ │ │ │ ├── dt_input.txt │ │ │ │ ├── dt_output.txt │ │ │ │ ├── dt_ref_img_gradient.txt │ │ │ │ ├── generate_sample_data.py │ │ │ │ ├── sample_data.h5 │ │ │ │ ├── sample_data_2_gzip.h5 │ │ │ │ ├── sample_data_list.txt │ │ │ │ ├── solver_data.h5 │ │ │ │ └── solver_data_list.txt │ │ │ ├── test_data_layer.cpp │ │ │ ├── test_data_transformer.cpp │ │ │ ├── test_db.cpp │ │ │ ├── test_deconvolution_layer.cpp │ │ │ ├── test_domain_transform_forward_only_layer.cpp │ │ │ ├── test_domain_transform_layer.cpp │ │ │ ├── test_dummy_data_layer.cpp │ │ │ ├── test_eltwise_layer.cpp │ │ │ ├── test_embed_layer.cpp │ │ │ ├── test_euclidean_loss_layer.cpp │ │ │ ├── test_filler.cpp │ │ │ ├── test_filter_layer.cpp │ │ │ ├── test_flatten_layer.cpp │ │ │ ├── test_gradient_based_solver.cpp │ │ │ ├── test_hdf5_output_layer.cpp │ │ │ ├── test_hdf5data_layer.cpp │ │ │ ├── test_hinge_loss_layer.cpp │ │ │ ├── test_im2col_kernel.cu │ │ │ ├── test_im2col_layer.cpp │ │ │ ├── test_image_data_layer.cpp │ │ │ ├── test_infogain_loss_layer.cpp │ │ │ ├── test_inner_product_layer.cpp │ │ │ ├── test_internal_thread.cpp │ │ │ ├── test_interp_layer.cpp │ │ │ ├── test_io.cpp │ │ │ ├── test_layer_factory.cpp │ │ │ ├── test_lrn_layer.cpp │ │ │ ├── test_math_functions.cpp │ │ │ ├── test_maxpool_dropout_layers.cpp │ │ │ ├── test_memory_data_layer.cpp │ │ │ ├── test_multinomial_logistic_loss_layer.cpp │ │ │ ├── test_mvn_layer.cpp │ │ │ ├── test_net.cpp │ │ │ ├── test_neuron_layer.cpp │ │ │ ├── test_platform.cpp │ │ │ ├── test_pooling_layer.cpp │ │ │ ├── test_power_layer.cpp │ │ │ ├── test_protobuf.cpp │ │ │ ├── test_random_number_generator.cpp │ │ │ ├── test_reduction_layer.cpp │ │ │ ├── test_reshape_layer.cpp │ │ │ ├── test_scale_layer.cpp │ │ │ ├── test_sigmoid_cross_entropy_loss_layer.cpp │ │ │ ├── test_slice_layer.cpp │ │ │ ├── test_softmax_layer.cpp │ │ │ ├── test_softmax_with_loss_layer.cpp │ │ │ ├── test_solver.cpp │ │ │ ├── test_solver_factory.cpp │ │ │ ├── test_spatial_product_layer.cpp │ │ │ ├── test_split_layer.cpp │ │ │ ├── test_spp_layer.cpp │ │ │ ├── test_stochastic_pooling.cpp │ │ │ ├── test_syncedmem.cpp │ │ │ ├── test_tanh_layer.cpp │ │ │ ├── test_threshold_layer.cpp │ │ │ ├── test_tile_layer.cpp │ │ │ ├── test_upgrade_proto.cpp │ │ │ └── test_util_blas.cpp │ │ └── util │ │ │ ├── benchmark.cpp │ │ │ ├── blocking_queue.cpp │ │ │ ├── confusion_matrix.cpp │ │ │ ├── cudnn.cpp │ │ │ ├── db.cpp │ │ │ ├── db_leveldb.cpp │ │ │ ├── db_lmdb.cpp │ │ │ ├── densecrf_pairwise.cpp │ │ │ ├── densecrf_util.cpp │ │ │ ├── hdf5.cpp │ │ │ ├── im2col.cpp │ │ │ ├── im2col.cu │ │ │ ├── insert_splits.cpp │ │ │ ├── interp.cpp │ │ │ ├── interp.cu │ │ │ ├── io.cpp │ │ │ ├── math_functions.cpp │ │ │ ├── math_functions.cu │ │ │ ├── matio_io.cpp │ │ │ ├── permutohedral.cpp │ │ │ ├── signal_handler.cpp │ │ │ └── upgrade_proto.cpp │ └── gtest │ │ ├── CMakeLists.txt │ │ ├── gtest-all.cpp │ │ ├── gtest.h │ │ └── gtest_main.cc └── tools │ ├── CMakeLists.txt │ ├── caffe.cpp │ ├── compute_image_mean.cpp │ ├── convert_imageset.cpp │ ├── device_query.cpp │ ├── extra │ ├── extract_seconds.py │ ├── launch_resize_and_crop_images.sh │ ├── parse_log.py │ ├── parse_log.sh │ ├── plot_log.gnuplot.example │ ├── plot_training_log.py.example │ ├── resize_and_crop_images.py │ └── summarize.py │ ├── extract_features.cpp │ ├── finetune_net.cpp │ ├── net_speed_benchmark.cpp │ ├── test_net.cpp │ ├── test_read_label.cpp │ ├── train_net.cpp │ ├── upgrade_net_proto_binary.cpp │ ├── upgrade_net_proto_text.cpp │ └── upgrade_solver_proto_text.cpp └── models_prototxts └── get_msrnet-vgg_model.sh /README.md: -------------------------------------------------------------------------------- 1 | # MSRNet 2 | * This is the sample code of saliency detection for 2017 cvpr paper [Instance-Level Salient Object Segmentation] 3 | by Guanbin Li, Yuan Xie, Liang Lin and Yizhou Yu 4 | * This code is tested on MATLAB 2014b on Ubuntu14.04 5 | * For more information, please visit our project page 6 | (http://www.sysu-hcp.net/instance-level-salient-object-segmentation) 7 | 8 | ## Contents ## 9 | This code includes 10 | - 'deeplab-caffe': the Caffe toolbox for Multiscale Refinement Network (MSRNet) 11 | - 'models_prototxts': pre-trained models and prototxts 12 | - 'code': codes to do testing 13 | - 'data': 14 | - a.imgs: source images to do saliency detection 15 | - b.pre: predicted results 16 | 17 | ## Usage Instructions ## 18 | Please follow the instructions below to run the code. 19 | - Compile the `Caffe` and `matcaffe` in the `deeplab-caffe` package. 20 | - Put your own images in the `data/imgs` directory 21 | - Download the pretrained MSRNet-VGG models by running the script 22 | ``` 23 | ./models_prototxts/get_msrnet-vgg_model.sh 24 | ``` 25 | - Generate saliency map by running the matlab code 26 | ``` 27 | ./code/demo.m 28 | ``` 29 | 30 | ## Citation ## 31 | If you find this useful, please cite our work as follows: 32 | ``` 33 | @inproceedings{MSRNet2017object, 34 | title={Instance-Level Salient Object Segmentation}, 35 | author={Guanbin Li, Yuan Xie, Liang Lin and Yizhou Yu}, 36 | booktitle={CVPR}, 37 | year={2017} 38 | } 39 | ``` 40 | Please contact "xiey39@mail2.sysu.edu.cn" if any questions with the code. 41 | 42 | -------------------------------------------------------------------------------- /data/imgs/0_0_272.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/data/imgs/0_0_272.jpg -------------------------------------------------------------------------------- /data/imgs/0_0_775.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/data/imgs/0_0_775.jpg -------------------------------------------------------------------------------- /deeplab-caffe/.gitignore: -------------------------------------------------------------------------------- 1 | ## General 2 | 3 | # Compiled Object files 4 | *.slo 5 | *.lo 6 | *.o 7 | *.cuo 8 | 9 | # Compiled Dynamic libraries 10 | *.so 11 | *.dylib 12 | 13 | # Compiled Static libraries 14 | *.lai 15 | *.la 16 | *.a 17 | 18 | # Compiled protocol buffers 19 | *.pb.h 20 | *.pb.cc 21 | *_pb2.py 22 | 23 | # Compiled python 24 | *.pyc 25 | 26 | # Compiled MATLAB 27 | *.mex* 28 | 29 | # IPython notebook checkpoints 30 | .ipynb_checkpoints 31 | 32 | # Editor temporaries 33 | *.swp 34 | *~ 35 | 36 | # Sublime Text settings 37 | *.sublime-workspace 38 | *.sublime-project 39 | 40 | # Eclipse Project settings 41 | *.*project 42 | .settings 43 | 44 | # QtCreator files 45 | *.user 46 | 47 | # PyCharm files 48 | .idea 49 | 50 | # OSX dir files 51 | .DS_Store 52 | 53 | ## Caffe 54 | 55 | # User's build configuration 56 | Makefile.config 57 | 58 | # Data and models are either 59 | # 1. reference, and not casually committed 60 | # 2. custom, and live on their own unless they're deliberated contributed 61 | data/* 62 | models/* 63 | *.caffemodel 64 | *.caffemodel.h5 65 | *.solverstate 66 | *.solverstate.h5 67 | *.binaryproto 68 | *leveldb 69 | *lmdb 70 | 71 | # build, distribute, and bins (+ python proto bindings) 72 | build 73 | .build_debug/* 74 | .build_release/* 75 | distribute/* 76 | *.testbin 77 | *.bin 78 | python/caffe/proto/ 79 | cmake_build 80 | .cmake_build 81 | 82 | # Generated documentation 83 | docs/_site 84 | docs/gathered 85 | _site 86 | doxygen 87 | docs/dev 88 | 89 | # LevelDB files 90 | *.sst 91 | *.ldb 92 | LOCK 93 | LOG* 94 | CURRENT 95 | MANIFEST-* 96 | -------------------------------------------------------------------------------- /deeplab-caffe/CONTRIBUTORS.md: -------------------------------------------------------------------------------- 1 | # Contributors 2 | 3 | Caffe is developed by a core set of BVLC members and the open-source community. 4 | 5 | We thank all of our [contributors](https://github.com/BVLC/caffe/graphs/contributors)! 6 | 7 | **For the detailed history of contributions** of a given file, try 8 | 9 | git blame file 10 | 11 | to see line-by-line credits and 12 | 13 | git log --follow file 14 | 15 | to see the change log even across renames and rewrites. 16 | 17 | Please refer to the [acknowledgements](http://caffe.berkeleyvision.org/#acknowledgements) on the Caffe site for further details. 18 | 19 | **Copyright** is held by the original contributor according to the versioning history; see LICENSE. 20 | -------------------------------------------------------------------------------- /deeplab-caffe/INSTALL.md: -------------------------------------------------------------------------------- 1 | # Installation 2 | 3 | See http://caffe.berkeleyvision.org/installation.html for the latest 4 | installation instructions. 5 | 6 | Check the users group in case you need help: 7 | https://groups.google.com/forum/#!forum/caffe-users 8 | -------------------------------------------------------------------------------- /deeplab-caffe/caffe.cloc: -------------------------------------------------------------------------------- 1 | Bourne Shell 2 | filter remove_matches ^\s*# 3 | filter remove_inline #.*$ 4 | extension sh 5 | script_exe sh 6 | C 7 | filter remove_matches ^\s*// 8 | filter call_regexp_common C 9 | filter remove_inline //.*$ 10 | extension c 11 | extension ec 12 | extension pgc 13 | C++ 14 | filter remove_matches ^\s*// 15 | filter remove_inline //.*$ 16 | filter call_regexp_common C 17 | extension C 18 | extension cc 19 | extension cpp 20 | extension cxx 21 | extension pcc 22 | C/C++ Header 23 | filter remove_matches ^\s*// 24 | filter call_regexp_common C 25 | filter remove_inline //.*$ 26 | extension H 27 | extension h 28 | extension hh 29 | extension hpp 30 | CUDA 31 | filter remove_matches ^\s*// 32 | filter remove_inline //.*$ 33 | filter call_regexp_common C 34 | extension cu 35 | Python 36 | filter remove_matches ^\s*# 37 | filter docstring_to_C 38 | filter call_regexp_common C 39 | filter remove_inline #.*$ 40 | extension py 41 | make 42 | filter remove_matches ^\s*# 43 | filter remove_inline #.*$ 44 | extension Gnumakefile 45 | extension Makefile 46 | extension am 47 | extension gnumakefile 48 | extension makefile 49 | filename Gnumakefile 50 | filename Makefile 51 | filename gnumakefile 52 | filename makefile 53 | script_exe make 54 | -------------------------------------------------------------------------------- /deeplab-caffe/cmake/Modules/FindLMDB.cmake: -------------------------------------------------------------------------------- 1 | # Try to find the LMBD libraries and headers 2 | # LMDB_FOUND - system has LMDB lib 3 | # LMDB_INCLUDE_DIR - the LMDB include directory 4 | # LMDB_LIBRARIES - Libraries needed to use LMDB 5 | 6 | # FindCWD based on FindGMP by: 7 | # Copyright (c) 2006, Laurent Montel, 8 | # 9 | # Redistribution and use is allowed according to the terms of the BSD license. 10 | 11 | # Adapted from FindCWD by: 12 | # Copyright 2013 Conrad Steenberg 13 | # Aug 31, 2013 14 | 15 | find_path(LMDB_INCLUDE_DIR NAMES lmdb.h PATHS "$ENV{LMDB_DIR}/include") 16 | find_library(LMDB_LIBRARIES NAMES lmdb PATHS "$ENV{LMDB_DIR}/lib" ) 17 | 18 | include(FindPackageHandleStandardArgs) 19 | find_package_handle_standard_args(LMDB DEFAULT_MSG LMDB_INCLUDE_DIR LMDB_LIBRARIES) 20 | 21 | if(LMDB_FOUND) 22 | message(STATUS "Found lmdb (include: ${LMDB_INCLUDE_DIR}, library: ${LMDB_LIBRARIES})") 23 | mark_as_advanced(LMDB_INCLUDE_DIR LMDB_LIBRARIES) 24 | 25 | caffe_parse_header(${LMDB_INCLUDE_DIR}/lmdb.h 26 | LMDB_VERSION_LINES MDB_VERSION_MAJOR MDB_VERSION_MINOR MDB_VERSION_PATCH) 27 | set(LMDB_VERSION "${MDB_VERSION_MAJOR}.${MDB_VERSION_MINOR}.${MDB_VERSION_PATCH}") 28 | endif() 29 | -------------------------------------------------------------------------------- /deeplab-caffe/cmake/Modules/FindSnappy.cmake: -------------------------------------------------------------------------------- 1 | # Find the Snappy libraries 2 | # 3 | # The following variables are optionally searched for defaults 4 | # Snappy_ROOT_DIR: Base directory where all Snappy components are found 5 | # 6 | # The following are set after configuration is done: 7 | # SNAPPY_FOUND 8 | # Snappy_INCLUDE_DIR 9 | # Snappy_LIBRARIES 10 | 11 | find_path(Snappy_INCLUDE_DIR NAMES snappy.h 12 | PATHS ${SNAPPY_ROOT_DIR} ${SNAPPY_ROOT_DIR}/include) 13 | 14 | find_library(Snappy_LIBRARIES NAMES snappy 15 | PATHS ${SNAPPY_ROOT_DIR} ${SNAPPY_ROOT_DIR}/lib) 16 | 17 | include(FindPackageHandleStandardArgs) 18 | find_package_handle_standard_args(Snappy DEFAULT_MSG Snappy_INCLUDE_DIR Snappy_LIBRARIES) 19 | 20 | if(SNAPPY_FOUND) 21 | message(STATUS "Found Snappy (include: ${Snappy_INCLUDE_DIR}, library: ${Snappy_LIBRARIES})") 22 | mark_as_advanced(Snappy_INCLUDE_DIR Snappy_LIBRARIES) 23 | 24 | caffe_parse_header(${Snappy_INCLUDE_DIR}/snappy-stubs-public.h 25 | SNAPPY_VERION_LINES SNAPPY_MAJOR SNAPPY_MINOR SNAPPY_PATCHLEVEL) 26 | set(Snappy_VERSION "${SNAPPY_MAJOR}.${SNAPPY_MINOR}.${SNAPPY_PATCHLEVEL}") 27 | endif() 28 | 29 | -------------------------------------------------------------------------------- /deeplab-caffe/cmake/Modules/FindvecLib.cmake: -------------------------------------------------------------------------------- 1 | # Find the vecLib libraries as part of Accelerate.framework or as standalon framework 2 | # 3 | # The following are set after configuration is done: 4 | # VECLIB_FOUND 5 | # vecLib_INCLUDE_DIR 6 | # vecLib_LINKER_LIBS 7 | 8 | 9 | if(NOT APPLE) 10 | return() 11 | endif() 12 | 13 | set(__veclib_include_suffix "Frameworks/vecLib.framework/Versions/Current/Headers") 14 | 15 | find_path(vecLib_INCLUDE_DIR vecLib.h 16 | DOC "vecLib include directory" 17 | PATHS /System/Library/${__veclib_include_suffix} 18 | /System/Library/Frameworks/Accelerate.framework/Versions/Current/${__veclib_include_suffix} 19 | /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX10.9.sdk/System/Library/Frameworks/Accelerate.framework/Versions/Current/Frameworks/vecLib.framework/Headers/) 20 | 21 | include(FindPackageHandleStandardArgs) 22 | find_package_handle_standard_args(vecLib DEFAULT_MSG vecLib_INCLUDE_DIR) 23 | 24 | if(VECLIB_FOUND) 25 | if(vecLib_INCLUDE_DIR MATCHES "^/System/Library/Frameworks/vecLib.framework.*") 26 | set(vecLib_LINKER_LIBS -lcblas "-framework vecLib") 27 | message(STATUS "Found standalone vecLib.framework") 28 | else() 29 | set(vecLib_LINKER_LIBS -lcblas "-framework Accelerate") 30 | message(STATUS "Found vecLib as part of Accelerate.framework") 31 | endif() 32 | 33 | mark_as_advanced(vecLib_INCLUDE_DIR) 34 | endif() 35 | -------------------------------------------------------------------------------- /deeplab-caffe/cmake/Templates/CaffeConfigVersion.cmake.in: -------------------------------------------------------------------------------- 1 | set(PACKAGE_VERSION "@Caffe_VERSION@") 2 | 3 | # Check whether the requested PACKAGE_FIND_VERSION is compatible 4 | if("${PACKAGE_VERSION}" VERSION_LESS "${PACKAGE_FIND_VERSION}") 5 | set(PACKAGE_VERSION_COMPATIBLE FALSE) 6 | else() 7 | set(PACKAGE_VERSION_COMPATIBLE TRUE) 8 | if ("${PACKAGE_VERSION}" VERSION_EQUAL "${PACKAGE_FIND_VERSION}") 9 | set(PACKAGE_VERSION_EXACT TRUE) 10 | endif() 11 | endif() 12 | -------------------------------------------------------------------------------- /deeplab-caffe/cmake/Templates/caffe_config.h.in: -------------------------------------------------------------------------------- 1 | /* Sources directory */ 2 | #define SOURCE_FOLDER "${PROJECT_SOURCE_DIR}" 3 | 4 | /* Binaries directory */ 5 | #define BINARY_FOLDER "${PROJECT_BINARY_DIR}" 6 | 7 | /* NVIDA Cuda */ 8 | #cmakedefine HAVE_CUDA 9 | 10 | /* NVIDA cuDNN */ 11 | #cmakedefine HAVE_CUDNN 12 | #cmakedefine USE_CUDNN 13 | 14 | /* NVIDA cuDNN */ 15 | #cmakedefine CPU_ONLY 16 | 17 | /* Test device */ 18 | #define CUDA_TEST_DEVICE ${CUDA_TEST_DEVICE} 19 | 20 | /* Temporary (TODO: remove) */ 21 | #if 1 22 | #define CMAKE_SOURCE_DIR SOURCE_FOLDER "/src/" 23 | #define EXAMPLES_SOURCE_DIR BINARY_FOLDER "/examples/" 24 | #define CMAKE_EXT ".gen.cmake" 25 | #else 26 | #define CMAKE_SOURCE_DIR "src/" 27 | #define EXAMPLES_SOURCE_DIR "examples/" 28 | #define CMAKE_EXT "" 29 | #endif 30 | 31 | /* Matlab */ 32 | #cmakedefine HAVE_MATLAB 33 | 34 | /* IO libraries */ 35 | #cmakedefine USE_OPENCV 36 | #cmakedefine USE_LEVELDB 37 | #cmakedefine USE_LMDB 38 | #cmakedefine ALLOW_LMDB_NOLOCK 39 | -------------------------------------------------------------------------------- /deeplab-caffe/copyweights.m: -------------------------------------------------------------------------------- 1 | % set your mat caffe path 2 | matcaffePath = '/home/phoenix/deeplab/code-v2/matlab/'; 3 | addpath(matcaffePath) 4 | addpath(genpath('./')) 5 | 6 | % Set parameters for the CNN model 7 | param.protoFile='/home/phoenix/temp/train_train.prototxt'; 8 | param.modelFile='/home/phoenix/temp/refine4-up_addSuper_9000.caffemodel'; 9 | 10 | net = caffe.Net(param.protoFile, param.modelFile, 'train'); -------------------------------------------------------------------------------- /deeplab-caffe/densecrf/Makefile: -------------------------------------------------------------------------------- 1 | # update the path variables 2 | 3 | CC = g++ 4 | CFLAGS = -W -Wall -O2 5 | #CFLAGS = -W -Wall -g 6 | 7 | all: 8 | make clean 9 | make prog_test_densecrf 10 | make prog_refine_pascal 11 | make prog_refine_pascal_v4 12 | 13 | clean: 14 | rm -f *.a 15 | rm -f *.o 16 | rm -f prog_test_densecrf 17 | rm -f prog_refine_pascal 18 | rm -f prog_refine_pascal_v4 19 | 20 | libDenseCRF.a: libDenseCRF/bipartitedensecrf.cpp libDenseCRF/densecrf.cpp libDenseCRF/filter.cpp libDenseCRF/permutohedral.cpp libDenseCRF/util.cpp libDenseCRF/densecrf.h libDenseCRF/fastmath.h libDenseCRF/permutohedral.h libDenseCRF/sse_defs.h libDenseCRF/util.h 21 | $(CC) libDenseCRF/bipartitedensecrf.cpp libDenseCRF/densecrf.cpp libDenseCRF/filter.cpp libDenseCRF/permutohedral.cpp libDenseCRF/util.cpp -c $(CFLAGS) 22 | ar rcs libDenseCRF.a bipartitedensecrf.o densecrf.o filter.o permutohedral.o util.o 23 | 24 | prog_test_densecrf: test_densecrf/simple_dense_inference.cpp libDenseCRF.a 25 | $(CC) test_densecrf/simple_dense_inference.cpp -o prog_test_densecrf $(CFLAGS) -L. -lDenseCRF 26 | 27 | prog_refine_pascal: refine_pascal/dense_inference.cpp refine_pascal/dense_inference.h util/Timer.h libDenseCRF.a 28 | $(CC) refine_pascal/dense_inference.cpp -o prog_refine_pascal $(CFLAGS) -L. -lDenseCRF -I./refine_pascal/ -I./util/ 29 | 30 | prog_refine_pascal_v4: refine_pascal_v4/dense_inference.cpp util/Timer.h libDenseCRF.a 31 | $(CC) refine_pascal_v4/dense_inference.cpp -o prog_refine_pascal_v4 $(CFLAGS) -L. -lDenseCRF -lmatio -lhdf5 -I./util/ 32 | 33 | -------------------------------------------------------------------------------- /deeplab-caffe/densecrf/README.md: -------------------------------------------------------------------------------- 1 | ## DenseCRF 2 | 3 | ### Code 4 | 5 | The code is modified from the publicly available code by Philipp Krähenbühl and Vladlen Koltun. 6 | See their project [website](http://www.philkr.net/home/densecrf) for more information 7 | 8 | If you also use this part of code, please cite their [paper](http://googledrive.com/host/0B6qziMs8hVGieFg0UzE0WmZaOW8/papers/densecrf.pdf): 9 | Efficient Inference in Fully Connected CRFs with Gaussian Edge Potentials, Philipp Krähenbühl and Vladlen Koltun, NIPS 2011. 10 | 11 | ### How to compile the code 12 | 13 | Linux: 14 | 15 | Run make command (modify Makefile if needed). 16 | 17 | Please see run_densecrf.sh for examples of input arguments or see the dense_inference.cpp. 18 | 19 | ### Caffe wrapper 20 | 21 | We have also provided a wrapper for Philipp's implementation in Caffe (see the layer densecrf_layer.cpp) -------------------------------------------------------------------------------- /deeplab-caffe/densecrf/my_script/LoadBinFile.m: -------------------------------------------------------------------------------- 1 | function out = LoadBinFile(fn, type) 2 | % load binary file 3 | 4 | fid = fopen(fn, 'rb'); 5 | 6 | row = fread(fid, 1, 'int32'); 7 | col = fread(fid, 1, 'int32'); 8 | channel = fread(fid, 1, 'int32'); 9 | 10 | numel = row * col * channel; 11 | 12 | if strcmp(type, 'double') 13 | out = fread(fid, numel, 'double'); 14 | out = double(out); 15 | elseif strcmp(type, 'single') || strcmp(type, 'float') 16 | out = fread(fid, numel, 'single'); 17 | out = single(out); 18 | elseif strcmp(type, 'uint8') 19 | out = fread(fid, numel, 'uint8'); 20 | out = uint8(out); 21 | elseif strcmp(type, 'int16') 22 | out = fread(fid, numel, 'int16'); 23 | out = int16(out); 24 | elseif strcmp(type, 'int32') 25 | out = fread(fid, numel, 'int32'); 26 | out = int32(out); 27 | else 28 | error('wrong type') 29 | end 30 | 31 | out = reshape(out, [row, col, channel]); 32 | 33 | fclose(fid); -------------------------------------------------------------------------------- /deeplab-caffe/densecrf/my_script/My_GetDenseCRFResult.m: -------------------------------------------------------------------------------- 1 | % compute the densecrf result (.bin) to png 2 | % 3 | function My_GetDenseCRFResult(stage) 4 | addpath('/home/phoenix/deeplab/code-v2/matlab/my_script'); 5 | %SetupEnv; 6 | load('./pascal_seg_colormap.mat'); 7 | 8 | map_folder=['/home/phoenix/Dataset/MSRA-B/SaliencyMap/' stage '_crf_bin']; 9 | map_dir = dir(fullfile(map_folder, '*bin')); 10 | save_result_folder=['/home/phoenix/Dataset/MSRA-B/SaliencyMap/' stage '_crf_png']; 11 | 12 | fprintf(1,' saving to %s\n', save_result_folder); 13 | 14 | if ~exist(save_result_folder, 'dir') 15 | mkdir(save_result_folder); 16 | end 17 | 18 | for i = 1 : numel(map_dir) 19 | fprintf(1, 'processing %d (%d)...\n', i, numel(map_dir)); 20 | map = LoadBinFile(fullfile(map_folder, map_dir(i).name), 'int16'); 21 | 22 | img_fn = map_dir(i).name(1:end-4); 23 | %imwrite(uint8(map), colormap, fullfile(save_result_folder, [img_fn, '.png'])); 24 | imwrite(double(map), fullfile(save_result_folder, [img_fn, '.png'])); 25 | end 26 | end 27 | 28 | -------------------------------------------------------------------------------- /deeplab-caffe/densecrf/my_script/SaveBinFile.m: -------------------------------------------------------------------------------- 1 | function SaveBinFile(data, fn, type) 2 | % save as binary file 3 | 4 | fid = fopen(fn, 'wb'); 5 | 6 | row = size(data, 1); 7 | col = size(data, 2); 8 | channel = size(data, 3); 9 | 10 | fwrite(fid, row, 'int32'); 11 | fwrite(fid, col, 'int32'); 12 | fwrite(fid, channel, 'int32'); 13 | 14 | if strcmp(type, 'double') 15 | fwrite(fid, data(:), 'double'); 16 | elseif strcmp(type, 'single') || strcmp(type, 'float') 17 | fwrite(fid, data(:), 'single'); 18 | elseif strcmp(type, 'uint8') 19 | fwrite(fid, data(:), 'uint8'); 20 | elseif strcmp(type, 'int32') 21 | fwrite(fid, data(:), 'int32'); 22 | else 23 | error('wrong type') 24 | end 25 | 26 | fclose(fid); -------------------------------------------------------------------------------- /deeplab-caffe/densecrf/my_script/SaveJpgToBin.m: -------------------------------------------------------------------------------- 1 | % save jpg images as bin file for cpp 2 | % 3 | 4 | img_folder = '../img'; 5 | save_folder = '../img_bin'; 6 | 7 | if ~exist(save_folder, 'dir') 8 | mkdir(save_folder); 9 | end 10 | 11 | img_dir = dir(fullfile(img_folder, '*.jpg')); 12 | 13 | for i = 1 : numel(img_dir) 14 | img = imread(fullfile(img_folder, img_dir(i).name)); 15 | 16 | img_fn = img_dir(i).name(1:end-4); 17 | save_fn = fullfile(save_folder, [img_fn, '.bin']); 18 | 19 | SaveBinFile(img, save_fn, 'uint8'); 20 | end 21 | -------------------------------------------------------------------------------- /deeplab-caffe/densecrf/my_script/SaveJpgToPPM.m: -------------------------------------------------------------------------------- 1 | % save jpg images as bin file for cpp 2 | % 3 | is_server = 1; 4 | 5 | dataset = 'PASCALContourData';%'msra' %'coco', 'voc2012','msra' 6 | 7 | if is_server 8 | if strcmp(dataset, 'voc2012') 9 | img_folder = '/rmt/data/pascal/VOCdevkit/VOC2012/JPEGImages' 10 | save_folder = '/rmt/data/pascal/VOCdevkit/VOC2012/PPMImages'; 11 | elseif strcmp(dataset, 'coco') 12 | img_folder = '/rmt/data/coco/JPEGImages'; 13 | save_folder = '/rmt/data/coco/PPMImages'; 14 | elseif strcmp(dataset,'msra') 15 | img_folder = '/home/phoenix/densecrf/examples/saliency/pre'%'/home/phoenix/deeplab/rmt/data/msra/image' 16 | save_folder = '/home/phoenix/densecrf/examples/saliency'%'/home/phoenix/deeplab/rmt/data/msra/PPMImages'; 17 | elseif strcmp(dataset,'PASCALContourData') 18 | img_folder = '/home/phoenix/Dataset/PASCALContourData/JPEGImages'%'/home/phoenix/deeplab/rmt/data/msra/image' 19 | save_folder = '/home/phoenix/Dataset/PASCALContourData/PPMImages' 20 | end 21 | else 22 | img_folder = '../img'; 23 | save_folder = '../img_ppm'; 24 | end 25 | 26 | if ~exist(save_folder, 'dir') 27 | mkdir(save_folder); 28 | end 29 | 30 | img_dir = dir(fullfile(img_folder, '*.jpg')); 31 | disp(numel(img_dir)); 32 | for i = 1 : numel(img_dir) 33 | fprintf(1, 'processing %d (%d)...\n', i, numel(img_dir)); 34 | img = imread(fullfile(img_folder, img_dir(i).name)); 35 | 36 | img_fn = img_dir(i).name(1:end-4); 37 | save_fn = fullfile(save_folder, [img_fn, '.ppm']); 38 | 39 | %img=img*255; 40 | imwrite(img, save_fn); 41 | end 42 | 43 | -------------------------------------------------------------------------------- /deeplab-caffe/densecrf/my_script/SaveMatAsBin.m: -------------------------------------------------------------------------------- 1 | % save mat score maps as bin file for cpp 2 | % 3 | %addpath('/rmt/work/deeplabel/code/matlab/my_script'); 4 | %SetupEnv; 5 | 6 | %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 7 | % You do not need to chage values below 8 | %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 9 | 10 | 11 | img_folder = '/home/phoenix/Dataset/MSRA-B/image'; 12 | mat_folder = '/home/phoenix/Dataset/MSRA-B/SaliencyMap/mean1_mat'; 13 | save_folder = '/home/phoenix/Dataset/MSRA-B/SaliencyMap/mean1_mat_bin'; 14 | 15 | if ~exist(save_folder, 'dir') 16 | mkdir(save_folder); 17 | end 18 | 19 | fprintf(1, 'Saving to %s\n', save_folder); 20 | 21 | mat_dir = dir(fullfile(mat_folder, '*.mat')); 22 | 23 | for i = 1 : numel(mat_dir) 24 | fprintf(1, 'processing %d (%d)...\n', i, numel(mat_dir)); 25 | data = load(fullfile(mat_folder, mat_dir(i).name)); 26 | data = data.data; 27 | % data = permute(data, [2 1 3]); 28 | %Transform data to probability 29 | data = exp(data); 30 | data = bsxfun(@rdivide, data, sum(data, 3)); 31 | 32 | img_fn = mat_dir(i).name(1:end-4); 33 | img_fn = strrep(img_fn, '_blob_0', ''); 34 | img = imread(fullfile(img_folder, [img_fn, '.jpg'])); 35 | img_row = size(img, 1); 36 | img_col = size(img, 2); 37 | 38 | data = data(1:img_row, 1:img_col, :); 39 | 40 | save_fn = fullfile(save_folder, [img_fn, '.bin']); 41 | 42 | SaveBinFile(data, save_fn, 'float'); 43 | end 44 | 45 | -------------------------------------------------------------------------------- /deeplab-caffe/densecrf/my_script/SavePngToPPM.m: -------------------------------------------------------------------------------- 1 | % save jpg images as bin file for cpp 2 | % 3 | is_server = 1; 4 | 5 | dataset = 'msra'; %'coco', 'voc2012','msra' 6 | 7 | if is_server 8 | if strcmp(dataset, 'voc2012') 9 | img_folder = '/rmt/data/pascal/VOCdevkit/VOC2012/JPEGImages' 10 | save_folder = '/rmt/data/pascal/VOCdevkit/VOC2012/PPMImages'; 11 | elseif strcmp(dataset, 'coco') 12 | img_folder = '/rmt/data/coco/JPEGImages'; 13 | save_folder = '/rmt/data/coco/PPMImages'; 14 | elseif strcmp(dataset,'msra') 15 | img_folder = '/home/phoenix/deeplab/rmt/work/deeplabel/exper/msra/res/features/deeplab_v2_vgg16_refine_full/val/fc8/post_none/results/Saliency/comp6_val_cls' 16 | save_folder = '/home/phoenix/deeplab/rmt/work/deeplabel/exper/msra/res/features/deeplab_v2_vgg16_refine_full/val/fc8/post_none/results/Saliency/ppm'; 17 | end 18 | else 19 | img_folder = '../img'; 20 | save_folder = '../img_ppm'; 21 | end 22 | 23 | if ~exist(save_folder, 'dir') 24 | mkdir(save_folder); 25 | end 26 | 27 | img_dir = dir(fullfile(img_folder, '*.png')); 28 | disp(numel(img_dir)); 29 | for i = 1 : numel(img_dir) 30 | fprintf(1, 'processing %d (%d)...\n', i, numel(img_dir)); 31 | img = imread(fullfile(img_folder, img_dir(i).name)); 32 | 33 | img_fn = img_dir(i).name(1:end-4); 34 | save_fn = fullfile(save_folder, [img_fn, '.ppm']); 35 | 36 | imwrite(img, save_fn); 37 | end 38 | 39 | -------------------------------------------------------------------------------- /deeplab-caffe/densecrf/my_script/pascal_seg_colormap.mat: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/densecrf/my_script/pascal_seg_colormap.mat -------------------------------------------------------------------------------- /deeplab-caffe/densecrf/prog_refine_pascal: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/densecrf/prog_refine_pascal -------------------------------------------------------------------------------- /deeplab-caffe/densecrf/prog_refine_pascal_v4: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/densecrf/prog_refine_pascal_v4 -------------------------------------------------------------------------------- /deeplab-caffe/densecrf/prog_test_densecrf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/densecrf/prog_test_densecrf -------------------------------------------------------------------------------- /deeplab-caffe/densecrf/refine_pascal/dense_inference.h: -------------------------------------------------------------------------------- 1 | #ifndef _DENSE_INFERENCE_H 2 | #define _DENSE_INFERENCE_H 3 | 4 | #include 5 | #include 6 | #include 7 | #include 8 | 9 | 10 | 11 | 12 | #endif 13 | -------------------------------------------------------------------------------- /deeplab-caffe/densecrf_float/Makefile: -------------------------------------------------------------------------------- 1 | # update the path variables 2 | 3 | CC = g++ 4 | CFLAGS = -W -Wall -O2 5 | #CFLAGS = -W -Wall -g 6 | 7 | all: 8 | make clean 9 | make prog_test_densecrf 10 | make prog_refine_pascal 11 | make prog_refine_pascal_v4 12 | 13 | clean: 14 | rm -f *.a 15 | rm -f *.o 16 | rm -f prog_test_densecrf 17 | rm -f prog_refine_pascal 18 | rm -f prog_refine_pascal_v4 19 | 20 | libDenseCRF.a: libDenseCRF/bipartitedensecrf.cpp libDenseCRF/densecrf.cpp libDenseCRF/filter.cpp libDenseCRF/permutohedral.cpp libDenseCRF/util.cpp libDenseCRF/densecrf.h libDenseCRF/fastmath.h libDenseCRF/permutohedral.h libDenseCRF/sse_defs.h libDenseCRF/util.h 21 | $(CC) libDenseCRF/bipartitedensecrf.cpp libDenseCRF/densecrf.cpp libDenseCRF/filter.cpp libDenseCRF/permutohedral.cpp libDenseCRF/util.cpp -c $(CFLAGS) 22 | ar rcs libDenseCRF.a bipartitedensecrf.o densecrf.o filter.o permutohedral.o util.o 23 | 24 | prog_test_densecrf: test_densecrf/simple_dense_inference.cpp libDenseCRF.a 25 | $(CC) test_densecrf/simple_dense_inference.cpp -o prog_test_densecrf $(CFLAGS) -L. -lDenseCRF 26 | 27 | prog_refine_pascal: refine_pascal/dense_inference.cpp refine_pascal/dense_inference.h util/Timer.h libDenseCRF.a 28 | $(CC) refine_pascal/dense_inference.cpp -o prog_refine_pascal $(CFLAGS) -L. -lDenseCRF -I./refine_pascal/ -I./util/ 29 | 30 | prog_refine_pascal_v4: refine_pascal_v4/dense_inference.cpp util/Timer.h libDenseCRF.a 31 | $(CC) refine_pascal_v4/dense_inference.cpp -o prog_refine_pascal_v4 $(CFLAGS) -L. -lDenseCRF -lmatio -lhdf5 -I./util/ 32 | 33 | -------------------------------------------------------------------------------- /deeplab-caffe/densecrf_float/README.md: -------------------------------------------------------------------------------- 1 | ## DenseCRF 2 | 3 | ### Code 4 | 5 | The code is modified from the publicly available code by Philipp Krähenbühl and Vladlen Koltun. 6 | See their project [website](http://www.philkr.net/home/densecrf) for more information 7 | 8 | If you also use this part of code, please cite their [paper](http://googledrive.com/host/0B6qziMs8hVGieFg0UzE0WmZaOW8/papers/densecrf.pdf): 9 | Efficient Inference in Fully Connected CRFs with Gaussian Edge Potentials, Philipp Krähenbühl and Vladlen Koltun, NIPS 2011. 10 | 11 | ### How to compile the code 12 | 13 | Linux: 14 | 15 | Run make command (modify Makefile if needed). 16 | 17 | Please see run_densecrf.sh for examples of input arguments or see the dense_inference.cpp. 18 | 19 | ### Caffe wrapper 20 | 21 | We have also provided a wrapper for Philipp's implementation in Caffe (see the layer densecrf_layer.cpp) -------------------------------------------------------------------------------- /deeplab-caffe/densecrf_float/my_script/LoadBinFile.m: -------------------------------------------------------------------------------- 1 | function out = LoadBinFile(fn, type) 2 | % load binary file 3 | 4 | fid = fopen(fn, 'rb'); 5 | 6 | row = fread(fid, 1, 'int32'); 7 | col = fread(fid, 1, 'int32'); 8 | channel = fread(fid, 1, 'int32'); 9 | 10 | numel = row * col * channel; 11 | 12 | if strcmp(type, 'double') 13 | out = fread(fid, numel, 'double'); 14 | out = double(out); 15 | elseif strcmp(type, 'single') || strcmp(type, 'float') 16 | out = fread(fid, numel, 'single'); 17 | out = single(out); 18 | elseif strcmp(type, 'uint8') 19 | out = fread(fid, numel, 'uint8'); 20 | out = uint8(out); 21 | elseif strcmp(type, 'int16') 22 | out = fread(fid, numel, 'int16'); 23 | out = int16(out); 24 | elseif strcmp(type, 'int32') 25 | out = fread(fid, numel, 'int32'); 26 | out = int32(out); 27 | else 28 | error('wrong type') 29 | end 30 | 31 | out = reshape(out, [row, col, channel]); 32 | 33 | fclose(fid); -------------------------------------------------------------------------------- /deeplab-caffe/densecrf_float/my_script/SaveBinFile.m: -------------------------------------------------------------------------------- 1 | function SaveBinFile(data, fn, type) 2 | % save as binary file 3 | 4 | fid = fopen(fn, 'wb'); 5 | 6 | row = size(data, 1); 7 | col = size(data, 2); 8 | channel = size(data, 3); 9 | 10 | fwrite(fid, row, 'int32'); 11 | fwrite(fid, col, 'int32'); 12 | fwrite(fid, channel, 'int32'); 13 | 14 | if strcmp(type, 'double') 15 | fwrite(fid, data(:), 'double'); 16 | elseif strcmp(type, 'single') || strcmp(type, 'float') 17 | fwrite(fid, data(:), 'single'); 18 | elseif strcmp(type, 'uint8') 19 | fwrite(fid, data(:), 'uint8'); 20 | elseif strcmp(type, 'int32') 21 | fwrite(fid, data(:), 'int32'); 22 | else 23 | error('wrong type') 24 | end 25 | 26 | fclose(fid); -------------------------------------------------------------------------------- /deeplab-caffe/densecrf_float/my_script/SaveJpgToBin.m: -------------------------------------------------------------------------------- 1 | % save jpg images as bin file for cpp 2 | % 3 | 4 | img_folder = '../img'; 5 | save_folder = '../img_bin'; 6 | 7 | if ~exist(save_folder, 'dir') 8 | mkdir(save_folder); 9 | end 10 | 11 | img_dir = dir(fullfile(img_folder, '*.jpg')); 12 | 13 | for i = 1 : numel(img_dir) 14 | img = imread(fullfile(img_folder, img_dir(i).name)); 15 | 16 | img_fn = img_dir(i).name(1:end-4); 17 | save_fn = fullfile(save_folder, [img_fn, '.bin']); 18 | 19 | SaveBinFile(img, save_fn, 'uint8'); 20 | end 21 | -------------------------------------------------------------------------------- /deeplab-caffe/densecrf_float/my_script/SaveJpgToPPM.m: -------------------------------------------------------------------------------- 1 | % save jpg images as bin file for cpp 2 | % 3 | is_server = 1; 4 | 5 | %dataset = 'coco'; %'coco', 'voc2012' 6 | dataset = 'voc2012'; 7 | 8 | if is_server 9 | if strcmp(dataset, 'voc2012') 10 | img_folder = '~/FCN_Saliency/deeplab/dataset/rmt/data/pascal/VOCdevkit/VOC2012/JPEGImages' 11 | save_folder = '~/FCN_Saliency/deeplab/dataset/rmt/data/pascal/VOCdevkit/VOC2012/PPMImages'; 12 | elseif strcmp(dataset, 'coco') 13 | img_folder = '~/FCN_Saliency/deeplab/dataset/rmt/data/pascal/JPEGImages'; 14 | save_folder = '~/FCN_Saliency/deeplab/dataset/rmt/data/pascal/PPMImages'; 15 | end 16 | else 17 | img_folder = '../img'; 18 | save_folder = '../img_ppm'; 19 | end 20 | 21 | if ~exist(save_folder, 'dir') 22 | mkdir(save_folder); 23 | end 24 | 25 | img_dir = dir(fullfile(img_folder, '*.jpg')); 26 | 27 | for i = 1 : numel(img_dir) 28 | fprintf(1, 'processing %d (%d)...\n', i, numel(img_dir)); 29 | img = imread(fullfile(img_folder, img_dir(i).name)); 30 | 31 | img_fn = img_dir(i).name(1:end-4); 32 | save_fn = fullfile(save_folder, [img_fn, '.ppm']); 33 | 34 | imwrite(img, save_fn); 35 | end 36 | 37 | -------------------------------------------------------------------------------- /deeplab-caffe/densecrf_float/my_script/bin2segimg.m: -------------------------------------------------------------------------------- 1 | 2 | 3 | %map_folder = '~/FCN_Saliency/deeplab/voc12/res/features/vgg128_nout/val/fc8/post_densecrf_w5_xstd50_posw3_poxsd50/'; 4 | map_folder = '~/FCN_Saliency/Deeplab_Saliency/dataset/msra/features2/vgg128_noup/test/floatcrf/'; 5 | 6 | map_dir = dir(fullfile(map_folder, '*.bin')); 7 | 8 | save_result_folder = '~/FCN_Saliency/Deeplab_Saliency/dataset/msra/features2/vgg128_noup/test/floatcrf/'; 9 | 10 | if ~exist(save_result_folder, 'dir') 11 | mkdir(save_result_folder); 12 | end 13 | 14 | for i = 1 : numel(map_dir) 15 | fprintf(1, 'processing %d (%d)...\n', i, numel(map_dir)); 16 | map = LoadBinFile(fullfile(map_folder, map_dir(i).name), 'float'); 17 | 18 | img_fn = map_dir(i).name(1:end-4); 19 | imwrite(uint8(map), colormap, fullfile(save_result_folder, [img_fn, '.png'])); 20 | end 21 | -------------------------------------------------------------------------------- /deeplab-caffe/densecrf_float/my_script/pascal_seg_colormap.mat: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/densecrf_float/my_script/pascal_seg_colormap.mat -------------------------------------------------------------------------------- /deeplab-caffe/densecrf_float/prog_refine_pascal: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/densecrf_float/prog_refine_pascal -------------------------------------------------------------------------------- /deeplab-caffe/densecrf_float/prog_refine_pascal_v4: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/densecrf_float/prog_refine_pascal_v4 -------------------------------------------------------------------------------- /deeplab-caffe/densecrf_float/prog_test_densecrf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/densecrf_float/prog_test_densecrf -------------------------------------------------------------------------------- /deeplab-caffe/densecrf_float/refine_pascal/dense_inference.h: -------------------------------------------------------------------------------- 1 | #ifndef _DENSE_INFERENCE_H 2 | #define _DENSE_INFERENCE_H 3 | 4 | #include 5 | #include 6 | #include 7 | #include 8 | 9 | 10 | 11 | 12 | #endif 13 | -------------------------------------------------------------------------------- /deeplab-caffe/docs/CNAME: -------------------------------------------------------------------------------- 1 | caffe.berkeleyvision.org 2 | -------------------------------------------------------------------------------- /deeplab-caffe/docs/README.md: -------------------------------------------------------------------------------- 1 | # Caffe Documentation 2 | 3 | To generate the documentation, run `$CAFFE_ROOT/scripts/build_docs.sh`. 4 | 5 | To push your changes to the documentation to the gh-pages branch of your or the BVLC repo, run `$CAFFE_ROOT/scripts/deploy_docs.sh `. 6 | -------------------------------------------------------------------------------- /deeplab-caffe/docs/_config.yml: -------------------------------------------------------------------------------- 1 | defaults: 2 | - 3 | scope: 4 | path: "" # an empty string here means all files in the project 5 | values: 6 | layout: "default" 7 | 8 | -------------------------------------------------------------------------------- /deeplab-caffe/docs/images/GitHub-Mark-64px.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/docs/images/GitHub-Mark-64px.png -------------------------------------------------------------------------------- /deeplab-caffe/docs/images/caffeine-icon.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/docs/images/caffeine-icon.png -------------------------------------------------------------------------------- /deeplab-caffe/docs/stylesheets/reset.css: -------------------------------------------------------------------------------- 1 | /* MeyerWeb Reset */ 2 | 3 | html, body, div, span, applet, object, iframe, 4 | h1, h2, h3, h4, h5, h6, p, blockquote, pre, 5 | a, abbr, acronym, address, big, cite, code, 6 | del, dfn, em, img, ins, kbd, q, s, samp, 7 | small, strike, strong, sub, sup, tt, var, 8 | b, u, i, center, 9 | dl, dt, dd, ol, ul, li, 10 | fieldset, form, label, legend, 11 | table, caption, tbody, tfoot, thead, tr, th, td, 12 | article, aside, canvas, details, embed, 13 | figure, figcaption, footer, header, hgroup, 14 | menu, nav, output, ruby, section, summary, 15 | time, mark, audio, video { 16 | margin: 0; 17 | padding: 0; 18 | border: 0; 19 | font: inherit; 20 | vertical-align: baseline; 21 | } 22 | -------------------------------------------------------------------------------- /deeplab-caffe/docs/tutorial/convolution.md: -------------------------------------------------------------------------------- 1 | --- 2 | title: Convolution 3 | --- 4 | # Caffeinated Convolution 5 | 6 | The Caffe strategy for convolution is to reduce the problem to matrix-matrix multiplication. 7 | This linear algebra computation is highly-tuned in BLAS libraries and efficiently computed on GPU devices. 8 | 9 | For more details read Yangqing's [Convolution in Caffe: a memo](https://github.com/Yangqing/caffe/wiki/Convolution-in-Caffe:-a-memo). 10 | 11 | As it turns out, this same reduction was independently explored in the context of conv. nets by 12 | 13 | > K. Chellapilla, S. Puri, P. Simard, et al. High performance convolutional neural networks for document processing. In Tenth International Workshop on Frontiers in Handwriting Recognition, 2006. 14 | -------------------------------------------------------------------------------- /deeplab-caffe/docs/tutorial/fig/.gitignore: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/docs/tutorial/fig/.gitignore -------------------------------------------------------------------------------- /deeplab-caffe/docs/tutorial/fig/backward.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/docs/tutorial/fig/backward.jpg -------------------------------------------------------------------------------- /deeplab-caffe/docs/tutorial/fig/forward.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/docs/tutorial/fig/forward.jpg -------------------------------------------------------------------------------- /deeplab-caffe/docs/tutorial/fig/forward_backward.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/docs/tutorial/fig/forward_backward.png -------------------------------------------------------------------------------- /deeplab-caffe/docs/tutorial/fig/layer.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/docs/tutorial/fig/layer.jpg -------------------------------------------------------------------------------- /deeplab-caffe/docs/tutorial/fig/logreg.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/docs/tutorial/fig/logreg.jpg -------------------------------------------------------------------------------- /deeplab-caffe/examples/CMakeLists.txt: -------------------------------------------------------------------------------- 1 | file(GLOB_RECURSE examples_srcs "${PROJECT_SOURCE_DIR}/examples/*.cpp") 2 | 3 | foreach(source_file ${examples_srcs}) 4 | # get file name 5 | get_filename_component(name ${source_file} NAME_WE) 6 | 7 | # get folder name 8 | get_filename_component(path ${source_file} PATH) 9 | get_filename_component(folder ${path} NAME_WE) 10 | 11 | add_executable(${name} ${source_file}) 12 | target_link_libraries(${name} ${Caffe_LINK}) 13 | caffe_default_properties(${name}) 14 | 15 | # set back RUNTIME_OUTPUT_DIRECTORY 16 | set_target_properties(${name} PROPERTIES 17 | RUNTIME_OUTPUT_DIRECTORY "${PROJECT_BINARY_DIR}/examples/${folder}") 18 | 19 | caffe_set_solution_folder(${name} examples) 20 | 21 | # install 22 | install(TARGETS ${name} DESTINATION bin) 23 | 24 | if(UNIX OR APPLE) 25 | # Funny command to make tutorials work 26 | # TODO: remove in future as soon as naming is standartaized everywhere 27 | set(__outname ${PROJECT_BINARY_DIR}/examples/${folder}/${name}${Caffe_POSTFIX}) 28 | add_custom_command(TARGET ${name} POST_BUILD 29 | COMMAND ln -sf "${__outname}" "${__outname}.bin") 30 | endif() 31 | endforeach() 32 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/cifar10/cifar10_full_sigmoid_solver.prototxt: -------------------------------------------------------------------------------- 1 | # reduce learning rate after 120 epochs (60000 iters) by factor 0f 10 2 | # then another factor of 10 after 10 more epochs (5000 iters) 3 | 4 | # The train/test net protocol buffer definition 5 | net: "examples/cifar10/cifar10_full_sigmoid_train_test.prototxt" 6 | # test_iter specifies how many forward passes the test should carry out. 7 | # In the case of CIFAR10, we have test batch size 100 and 100 test iterations, 8 | # covering the full 10,000 testing images. 9 | test_iter: 10 10 | # Carry out testing every 1000 training iterations. 11 | test_interval: 1000 12 | # The base learning rate, momentum and the weight decay of the network. 13 | base_lr: 0.001 14 | momentum: 0.9 15 | #weight_decay: 0.004 16 | # The learning rate policy 17 | lr_policy: "step" 18 | gamma: 1 19 | stepsize: 5000 20 | # Display every 200 iterations 21 | display: 100 22 | # The maximum number of iterations 23 | max_iter: 60000 24 | # snapshot intermediate results 25 | snapshot: 10000 26 | snapshot_prefix: "examples/cifar10_full_sigmoid" 27 | # solver mode: CPU or GPU 28 | solver_mode: GPU 29 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/cifar10/cifar10_full_sigmoid_solver_bn.prototxt: -------------------------------------------------------------------------------- 1 | # reduce learning rate after 120 epochs (60000 iters) by factor 0f 10 2 | # then another factor of 10 after 10 more epochs (5000 iters) 3 | 4 | # The train/test net protocol buffer definition 5 | net: "examples/cifar10/cifar10_full_sigmoid_train_test_bn.prototxt" 6 | # test_iter specifies how many forward passes the test should carry out. 7 | # In the case of CIFAR10, we have test batch size 100 and 100 test iterations, 8 | # covering the full 10,000 testing images. 9 | test_iter: 10 10 | # Carry out testing every 1000 training iterations. 11 | test_interval: 1000 12 | # The base learning rate, momentum and the weight decay of the network. 13 | base_lr: 0.001 14 | momentum: 0.9 15 | #weight_decay: 0.004 16 | # The learning rate policy 17 | lr_policy: "step" 18 | gamma: 1 19 | stepsize: 5000 20 | # Display every 200 iterations 21 | display: 100 22 | # The maximum number of iterations 23 | max_iter: 60000 24 | # snapshot intermediate results 25 | snapshot: 10000 26 | snapshot_prefix: "examples/cifar10_full_sigmoid_bn" 27 | # solver mode: CPU or GPU 28 | solver_mode: GPU 29 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/cifar10/cifar10_full_solver.prototxt: -------------------------------------------------------------------------------- 1 | # reduce learning rate after 120 epochs (60000 iters) by factor 0f 10 2 | # then another factor of 10 after 10 more epochs (5000 iters) 3 | 4 | # The train/test net protocol buffer definition 5 | net: "examples/cifar10/cifar10_full_train_test.prototxt" 6 | # test_iter specifies how many forward passes the test should carry out. 7 | # In the case of CIFAR10, we have test batch size 100 and 100 test iterations, 8 | # covering the full 10,000 testing images. 9 | test_iter: 100 10 | # Carry out testing every 1000 training iterations. 11 | test_interval: 1000 12 | # The base learning rate, momentum and the weight decay of the network. 13 | base_lr: 0.001 14 | momentum: 0.9 15 | weight_decay: 0.004 16 | # The learning rate policy 17 | lr_policy: "fixed" 18 | # Display every 200 iterations 19 | display: 200 20 | # The maximum number of iterations 21 | max_iter: 60000 22 | # snapshot intermediate results 23 | snapshot: 10000 24 | snapshot_format: HDF5 25 | snapshot_prefix: "examples/cifar10/cifar10_full" 26 | # solver mode: CPU or GPU 27 | solver_mode: GPU 28 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/cifar10/cifar10_full_solver_lr1.prototxt: -------------------------------------------------------------------------------- 1 | # reduce learning rate after 120 epochs (60000 iters) by factor 0f 10 2 | # then another factor of 10 after 10 more epochs (5000 iters) 3 | 4 | # The train/test net protocol buffer definition 5 | net: "examples/cifar10/cifar10_full_train_test.prototxt" 6 | # test_iter specifies how many forward passes the test should carry out. 7 | # In the case of CIFAR10, we have test batch size 100 and 100 test iterations, 8 | # covering the full 10,000 testing images. 9 | test_iter: 100 10 | # Carry out testing every 1000 training iterations. 11 | test_interval: 1000 12 | # The base learning rate, momentum and the weight decay of the network. 13 | base_lr: 0.0001 14 | momentum: 0.9 15 | weight_decay: 0.004 16 | # The learning rate policy 17 | lr_policy: "fixed" 18 | # Display every 200 iterations 19 | display: 200 20 | # The maximum number of iterations 21 | max_iter: 65000 22 | # snapshot intermediate results 23 | snapshot: 5000 24 | snapshot_format: HDF5 25 | snapshot_prefix: "examples/cifar10/cifar10_full" 26 | # solver mode: CPU or GPU 27 | solver_mode: GPU 28 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/cifar10/cifar10_full_solver_lr2.prototxt: -------------------------------------------------------------------------------- 1 | # reduce learning rate after 120 epochs (60000 iters) by factor 0f 10 2 | # then another factor of 10 after 10 more epochs (5000 iters) 3 | 4 | # The train/test net protocol buffer definition 5 | net: "examples/cifar10/cifar10_full_train_test.prototxt" 6 | # test_iter specifies how many forward passes the test should carry out. 7 | # In the case of CIFAR10, we have test batch size 100 and 100 test iterations, 8 | # covering the full 10,000 testing images. 9 | test_iter: 100 10 | # Carry out testing every 1000 training iterations. 11 | test_interval: 1000 12 | # The base learning rate, momentum and the weight decay of the network. 13 | base_lr: 0.00001 14 | momentum: 0.9 15 | weight_decay: 0.004 16 | # The learning rate policy 17 | lr_policy: "fixed" 18 | # Display every 200 iterations 19 | display: 200 20 | # The maximum number of iterations 21 | max_iter: 70000 22 | # snapshot intermediate results 23 | snapshot: 5000 24 | snapshot_format: HDF5 25 | snapshot_prefix: "examples/cifar10/cifar10_full" 26 | # solver mode: CPU or GPU 27 | solver_mode: GPU 28 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/cifar10/cifar10_quick_solver.prototxt: -------------------------------------------------------------------------------- 1 | # reduce the learning rate after 8 epochs (4000 iters) by a factor of 10 2 | 3 | # The train/test net protocol buffer definition 4 | net: "examples/cifar10/cifar10_quick_train_test.prototxt" 5 | # test_iter specifies how many forward passes the test should carry out. 6 | # In the case of MNIST, we have test batch size 100 and 100 test iterations, 7 | # covering the full 10,000 testing images. 8 | test_iter: 100 9 | # Carry out testing every 500 training iterations. 10 | test_interval: 500 11 | # The base learning rate, momentum and the weight decay of the network. 12 | base_lr: 0.001 13 | momentum: 0.9 14 | weight_decay: 0.004 15 | # The learning rate policy 16 | lr_policy: "fixed" 17 | # Display every 100 iterations 18 | display: 100 19 | # The maximum number of iterations 20 | max_iter: 4000 21 | # snapshot intermediate results 22 | snapshot: 4000 23 | snapshot_format: HDF5 24 | snapshot_prefix: "examples/cifar10/cifar10_quick" 25 | # solver mode: CPU or GPU 26 | solver_mode: GPU 27 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/cifar10/cifar10_quick_solver_lr1.prototxt: -------------------------------------------------------------------------------- 1 | # reduce the learning rate after 8 epochs (4000 iters) by a factor of 10 2 | 3 | # The train/test net protocol buffer definition 4 | net: "examples/cifar10/cifar10_quick_train_test.prototxt" 5 | # test_iter specifies how many forward passes the test should carry out. 6 | # In the case of MNIST, we have test batch size 100 and 100 test iterations, 7 | # covering the full 10,000 testing images. 8 | test_iter: 100 9 | # Carry out testing every 500 training iterations. 10 | test_interval: 500 11 | # The base learning rate, momentum and the weight decay of the network. 12 | base_lr: 0.0001 13 | momentum: 0.9 14 | weight_decay: 0.004 15 | # The learning rate policy 16 | lr_policy: "fixed" 17 | # Display every 100 iterations 18 | display: 100 19 | # The maximum number of iterations 20 | max_iter: 5000 21 | # snapshot intermediate results 22 | snapshot: 5000 23 | snapshot_format: HDF5 24 | snapshot_prefix: "examples/cifar10/cifar10_quick" 25 | # solver mode: CPU or GPU 26 | solver_mode: GPU 27 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/cifar10/create_cifar10.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env sh 2 | # This script converts the cifar data into leveldb format. 3 | 4 | EXAMPLE=examples/cifar10 5 | DATA=data/cifar10 6 | DBTYPE=lmdb 7 | 8 | echo "Creating $DBTYPE..." 9 | 10 | rm -rf $EXAMPLE/cifar10_train_$DBTYPE $EXAMPLE/cifar10_test_$DBTYPE 11 | 12 | ./build/examples/cifar10/convert_cifar_data.bin $DATA $EXAMPLE $DBTYPE 13 | 14 | echo "Computing image mean..." 15 | 16 | ./build/tools/compute_image_mean -backend=$DBTYPE \ 17 | $EXAMPLE/cifar10_train_$DBTYPE $EXAMPLE/mean.binaryproto 18 | 19 | echo "Done." 20 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/cifar10/train_full.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env sh 2 | 3 | TOOLS=./build/tools 4 | 5 | $TOOLS/caffe train \ 6 | --solver=examples/cifar10/cifar10_full_solver.prototxt 7 | 8 | # reduce learning rate by factor of 10 9 | $TOOLS/caffe train \ 10 | --solver=examples/cifar10/cifar10_full_solver_lr1.prototxt \ 11 | --snapshot=examples/cifar10/cifar10_full_iter_60000.solverstate.h5 12 | 13 | # reduce learning rate by factor of 10 14 | $TOOLS/caffe train \ 15 | --solver=examples/cifar10/cifar10_full_solver_lr2.prototxt \ 16 | --snapshot=examples/cifar10/cifar10_full_iter_65000.solverstate.h5 17 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/cifar10/train_full_sigmoid.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env sh 2 | 3 | TOOLS=./build/tools 4 | 5 | $TOOLS/caffe train \ 6 | --solver=examples/cifar10/cifar10_full_sigmoid_solver.prototxt 7 | 8 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/cifar10/train_full_sigmoid_bn.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env sh 2 | 3 | TOOLS=./build/tools 4 | 5 | $TOOLS/caffe train \ 6 | --solver=examples/cifar10/cifar10_full_sigmoid_solver_bn.prototxt 7 | 8 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/cifar10/train_quick.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env sh 2 | 3 | TOOLS=./build/tools 4 | 5 | $TOOLS/caffe train \ 6 | --solver=examples/cifar10/cifar10_quick_solver.prototxt 7 | 8 | # reduce learning rate by factor of 10 after 8 epochs 9 | $TOOLS/caffe train \ 10 | --solver=examples/cifar10/cifar10_quick_solver_lr1.prototxt \ 11 | --snapshot=examples/cifar10/cifar10_quick_iter_4000.solverstate.h5 12 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/finetune_flickr_style/flickr_style.csv.gz: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/examples/finetune_flickr_style/flickr_style.csv.gz -------------------------------------------------------------------------------- /deeplab-caffe/examples/finetune_flickr_style/style_names.txt: -------------------------------------------------------------------------------- 1 | Detailed 2 | Pastel 3 | Melancholy 4 | Noir 5 | HDR 6 | Vintage 7 | Long Exposure 8 | Horror 9 | Sunny 10 | Bright 11 | Hazy 12 | Bokeh 13 | Serene 14 | Texture 15 | Ethereal 16 | Macro 17 | Depth of Field 18 | Geometric Composition 19 | Minimal 20 | Romantic 21 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/finetune_pascal_detection/pascal_finetune_solver.prototxt: -------------------------------------------------------------------------------- 1 | net: "examples/finetune_pascal_detection/pascal_finetune_trainval_test.prototxt" 2 | test_iter: 100 3 | test_interval: 1000 4 | base_lr: 0.001 5 | lr_policy: "step" 6 | gamma: 0.1 7 | stepsize: 20000 8 | display: 20 9 | max_iter: 100000 10 | momentum: 0.9 11 | weight_decay: 0.0005 12 | snapshot: 10000 13 | snapshot_prefix: "examples/finetune_pascal_detection/pascal_det_finetune" 14 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/hdf5_classification/nonlinear_auto_test.prototxt: -------------------------------------------------------------------------------- 1 | layer { 2 | name: "data" 3 | type: "HDF5Data" 4 | top: "data" 5 | top: "label" 6 | hdf5_data_param { 7 | source: "examples/hdf5_classification/data/test.txt" 8 | batch_size: 10 9 | } 10 | } 11 | layer { 12 | name: "ip1" 13 | type: "InnerProduct" 14 | bottom: "data" 15 | top: "ip1" 16 | inner_product_param { 17 | num_output: 40 18 | weight_filler { 19 | type: "xavier" 20 | } 21 | } 22 | } 23 | layer { 24 | name: "relu1" 25 | type: "ReLU" 26 | bottom: "ip1" 27 | top: "ip1" 28 | } 29 | layer { 30 | name: "ip2" 31 | type: "InnerProduct" 32 | bottom: "ip1" 33 | top: "ip2" 34 | inner_product_param { 35 | num_output: 2 36 | weight_filler { 37 | type: "xavier" 38 | } 39 | } 40 | } 41 | layer { 42 | name: "accuracy" 43 | type: "Accuracy" 44 | bottom: "ip2" 45 | bottom: "label" 46 | top: "accuracy" 47 | } 48 | layer { 49 | name: "loss" 50 | type: "SoftmaxWithLoss" 51 | bottom: "ip2" 52 | bottom: "label" 53 | top: "loss" 54 | } 55 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/hdf5_classification/nonlinear_auto_train.prototxt: -------------------------------------------------------------------------------- 1 | layer { 2 | name: "data" 3 | type: "HDF5Data" 4 | top: "data" 5 | top: "label" 6 | hdf5_data_param { 7 | source: "examples/hdf5_classification/data/train.txt" 8 | batch_size: 10 9 | } 10 | } 11 | layer { 12 | name: "ip1" 13 | type: "InnerProduct" 14 | bottom: "data" 15 | top: "ip1" 16 | inner_product_param { 17 | num_output: 40 18 | weight_filler { 19 | type: "xavier" 20 | } 21 | } 22 | } 23 | layer { 24 | name: "relu1" 25 | type: "ReLU" 26 | bottom: "ip1" 27 | top: "ip1" 28 | } 29 | layer { 30 | name: "ip2" 31 | type: "InnerProduct" 32 | bottom: "ip1" 33 | top: "ip2" 34 | inner_product_param { 35 | num_output: 2 36 | weight_filler { 37 | type: "xavier" 38 | } 39 | } 40 | } 41 | layer { 42 | name: "accuracy" 43 | type: "Accuracy" 44 | bottom: "ip2" 45 | bottom: "label" 46 | top: "accuracy" 47 | } 48 | layer { 49 | name: "loss" 50 | type: "SoftmaxWithLoss" 51 | bottom: "ip2" 52 | bottom: "label" 53 | top: "loss" 54 | } 55 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/hdf5_classification/train_val.prototxt: -------------------------------------------------------------------------------- 1 | name: "LogisticRegressionNet" 2 | layer { 3 | name: "data" 4 | type: "HDF5Data" 5 | top: "data" 6 | top: "label" 7 | include { 8 | phase: TRAIN 9 | } 10 | hdf5_data_param { 11 | source: "examples/hdf5_classification/data/train.txt" 12 | batch_size: 10 13 | } 14 | } 15 | layer { 16 | name: "data" 17 | type: "HDF5Data" 18 | top: "data" 19 | top: "label" 20 | include { 21 | phase: TEST 22 | } 23 | hdf5_data_param { 24 | source: "examples/hdf5_classification/data/test.txt" 25 | batch_size: 10 26 | } 27 | } 28 | layer { 29 | name: "fc1" 30 | type: "InnerProduct" 31 | bottom: "data" 32 | top: "fc1" 33 | param { 34 | lr_mult: 1 35 | decay_mult: 1 36 | } 37 | param { 38 | lr_mult: 2 39 | decay_mult: 0 40 | } 41 | inner_product_param { 42 | num_output: 2 43 | weight_filler { 44 | type: "xavier" 45 | } 46 | bias_filler { 47 | type: "constant" 48 | value: 0 49 | } 50 | } 51 | } 52 | layer { 53 | name: "loss" 54 | type: "SoftmaxWithLoss" 55 | bottom: "fc1" 56 | bottom: "label" 57 | top: "loss" 58 | } 59 | layer { 60 | name: "accuracy" 61 | type: "Accuracy" 62 | bottom: "fc1" 63 | bottom: "label" 64 | top: "accuracy" 65 | include { 66 | phase: TEST 67 | } 68 | } 69 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/imagenet/make_imagenet_mean.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env sh 2 | # Compute the mean image from the imagenet training lmdb 3 | # N.B. this is available in data/ilsvrc12 4 | 5 | EXAMPLE=examples/imagenet 6 | DATA=data/ilsvrc12 7 | TOOLS=build/tools 8 | 9 | $TOOLS/compute_image_mean $EXAMPLE/ilsvrc12_train_lmdb \ 10 | $DATA/imagenet_mean.binaryproto 11 | 12 | echo "Done." 13 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/imagenet/resume_training.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env sh 2 | 3 | ./build/tools/caffe train \ 4 | --solver=models/bvlc_reference_caffenet/solver.prototxt \ 5 | --snapshot=models/bvlc_reference_caffenet/caffenet_train_10000.solverstate.h5 6 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/imagenet/train_caffenet.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env sh 2 | 3 | ./build/tools/caffe train \ 4 | --solver=models/bvlc_reference_caffenet/solver.prototxt 5 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/images/cat.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/examples/images/cat.jpg -------------------------------------------------------------------------------- /deeplab-caffe/examples/images/cat_gray.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/examples/images/cat_gray.jpg -------------------------------------------------------------------------------- /deeplab-caffe/examples/images/fish-bike.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/examples/images/fish-bike.jpg -------------------------------------------------------------------------------- /deeplab-caffe/examples/mnist/create_mnist.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env sh 2 | # This script converts the mnist data into lmdb/leveldb format, 3 | # depending on the value assigned to $BACKEND. 4 | 5 | EXAMPLE=examples/mnist 6 | DATA=data/mnist 7 | BUILD=build/examples/mnist 8 | 9 | BACKEND="lmdb" 10 | 11 | echo "Creating ${BACKEND}..." 12 | 13 | rm -rf $EXAMPLE/mnist_train_${BACKEND} 14 | rm -rf $EXAMPLE/mnist_test_${BACKEND} 15 | 16 | $BUILD/convert_mnist_data.bin $DATA/train-images-idx3-ubyte \ 17 | $DATA/train-labels-idx1-ubyte $EXAMPLE/mnist_train_${BACKEND} --backend=${BACKEND} 18 | $BUILD/convert_mnist_data.bin $DATA/t10k-images-idx3-ubyte \ 19 | $DATA/t10k-labels-idx1-ubyte $EXAMPLE/mnist_test_${BACKEND} --backend=${BACKEND} 20 | 21 | echo "Done." 22 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/mnist/lenet_adadelta_solver.prototxt: -------------------------------------------------------------------------------- 1 | # The train/test net protocol buffer definition 2 | net: "examples/mnist/lenet_train_test.prototxt" 3 | # test_iter specifies how many forward passes the test should carry out. 4 | # In the case of MNIST, we have test batch size 100 and 100 test iterations, 5 | # covering the full 10,000 testing images. 6 | test_iter: 100 7 | # Carry out testing every 500 training iterations. 8 | test_interval: 500 9 | # The base learning rate, momentum and the weight decay of the network. 10 | base_lr: 1.0 11 | lr_policy: "fixed" 12 | momentum: 0.95 13 | weight_decay: 0.0005 14 | # Display every 100 iterations 15 | display: 100 16 | # The maximum number of iterations 17 | max_iter: 10000 18 | # snapshot intermediate results 19 | snapshot: 5000 20 | snapshot_prefix: "examples/mnist/lenet_adadelta" 21 | # solver mode: CPU or GPU 22 | solver_mode: GPU 23 | type: "AdaDelta" 24 | delta: 1e-6 25 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/mnist/lenet_auto_solver.prototxt: -------------------------------------------------------------------------------- 1 | # The train/test net protocol buffer definition 2 | train_net: "mnist/lenet_auto_train.prototxt" 3 | test_net: "mnist/lenet_auto_test.prototxt" 4 | # test_iter specifies how many forward passes the test should carry out. 5 | # In the case of MNIST, we have test batch size 100 and 100 test iterations, 6 | # covering the full 10,000 testing images. 7 | test_iter: 100 8 | # Carry out testing every 500 training iterations. 9 | test_interval: 500 10 | # The base learning rate, momentum and the weight decay of the network. 11 | base_lr: 0.01 12 | momentum: 0.9 13 | weight_decay: 0.0005 14 | # The learning rate policy 15 | lr_policy: "inv" 16 | gamma: 0.0001 17 | power: 0.75 18 | # Display every 100 iterations 19 | display: 100 20 | # The maximum number of iterations 21 | max_iter: 10000 22 | # snapshot intermediate results 23 | snapshot: 5000 24 | snapshot_prefix: "mnist/lenet" 25 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/mnist/lenet_multistep_solver.prototxt: -------------------------------------------------------------------------------- 1 | # The train/test net protocol buffer definition 2 | net: "examples/mnist/lenet_train_test.prototxt" 3 | # test_iter specifies how many forward passes the test should carry out. 4 | # In the case of MNIST, we have test batch size 100 and 100 test iterations, 5 | # covering the full 10,000 testing images. 6 | test_iter: 100 7 | # Carry out testing every 500 training iterations. 8 | test_interval: 500 9 | # The base learning rate, momentum and the weight decay of the network. 10 | base_lr: 0.01 11 | momentum: 0.9 12 | weight_decay: 0.0005 13 | # The learning rate policy 14 | lr_policy: "multistep" 15 | gamma: 0.9 16 | stepvalue: 5000 17 | stepvalue: 7000 18 | stepvalue: 8000 19 | stepvalue: 9000 20 | stepvalue: 9500 21 | # Display every 100 iterations 22 | display: 100 23 | # The maximum number of iterations 24 | max_iter: 10000 25 | # snapshot intermediate results 26 | snapshot: 5000 27 | snapshot_prefix: "examples/mnist/lenet_multistep" 28 | # solver mode: CPU or GPU 29 | solver_mode: GPU 30 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/mnist/lenet_solver.prototxt: -------------------------------------------------------------------------------- 1 | # The train/test net protocol buffer definition 2 | net: "examples/mnist/lenet_train_test.prototxt" 3 | # test_iter specifies how many forward passes the test should carry out. 4 | # In the case of MNIST, we have test batch size 100 and 100 test iterations, 5 | # covering the full 10,000 testing images. 6 | test_iter: 100 7 | # Carry out testing every 500 training iterations. 8 | test_interval: 500 9 | # The base learning rate, momentum and the weight decay of the network. 10 | base_lr: 0.01 11 | momentum: 0.9 12 | weight_decay: 0.0005 13 | # The learning rate policy 14 | lr_policy: "inv" 15 | gamma: 0.0001 16 | power: 0.75 17 | # Display every 100 iterations 18 | display: 100 19 | # The maximum number of iterations 20 | max_iter: 10000 21 | # snapshot intermediate results 22 | snapshot: 5000 23 | snapshot_prefix: "examples/mnist/lenet" 24 | # solver mode: CPU or GPU 25 | solver_mode: GPU 26 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/mnist/lenet_solver_adam.prototxt: -------------------------------------------------------------------------------- 1 | # The train/test net protocol buffer definition 2 | # this follows "ADAM: A METHOD FOR STOCHASTIC OPTIMIZATION" 3 | net: "examples/mnist/lenet_train_test.prototxt" 4 | # test_iter specifies how many forward passes the test should carry out. 5 | # In the case of MNIST, we have test batch size 100 and 100 test iterations, 6 | # covering the full 10,000 testing images. 7 | test_iter: 100 8 | # Carry out testing every 500 training iterations. 9 | test_interval: 500 10 | # All parameters are from the cited paper above 11 | base_lr: 0.001 12 | momentum: 0.9 13 | momentum2: 0.999 14 | # since Adam dynamically changes the learning rate, we set the base learning 15 | # rate to a fixed value 16 | lr_policy: "fixed" 17 | # Display every 100 iterations 18 | display: 100 19 | # The maximum number of iterations 20 | max_iter: 10000 21 | # snapshot intermediate results 22 | snapshot: 5000 23 | snapshot_prefix: "examples/mnist/lenet" 24 | # solver mode: CPU or GPU 25 | type: "Adam" 26 | solver_mode: GPU 27 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/mnist/lenet_solver_rmsprop.prototxt: -------------------------------------------------------------------------------- 1 | # The train/test net protocol buffer definition 2 | net: "examples/mnist/lenet_train_test.prototxt" 3 | # test_iter specifies how many forward passes the test should carry out. 4 | # In the case of MNIST, we have test batch size 100 and 100 test iterations, 5 | # covering the full 10,000 testing images. 6 | test_iter: 100 7 | # Carry out testing every 500 training iterations. 8 | test_interval: 500 9 | # The base learning rate, momentum and the weight decay of the network. 10 | base_lr: 0.01 11 | momentum: 0.0 12 | weight_decay: 0.0005 13 | # The learning rate policy 14 | lr_policy: "inv" 15 | gamma: 0.0001 16 | power: 0.75 17 | # Display every 100 iterations 18 | display: 100 19 | # The maximum number of iterations 20 | max_iter: 10000 21 | # snapshot intermediate results 22 | snapshot: 5000 23 | snapshot_prefix: "examples/mnist/lenet_rmsprop" 24 | # solver mode: CPU or GPU 25 | solver_mode: GPU 26 | type: "RMSProp" 27 | rms_decay: 0.98 28 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/mnist/mnist_autoencoder_solver.prototxt: -------------------------------------------------------------------------------- 1 | net: "examples/mnist/mnist_autoencoder.prototxt" 2 | test_state: { stage: 'test-on-train' } 3 | test_iter: 500 4 | test_state: { stage: 'test-on-test' } 5 | test_iter: 100 6 | test_interval: 500 7 | test_compute_loss: true 8 | base_lr: 0.01 9 | lr_policy: "step" 10 | gamma: 0.1 11 | stepsize: 10000 12 | display: 100 13 | max_iter: 65000 14 | weight_decay: 0.0005 15 | snapshot: 10000 16 | snapshot_prefix: "examples/mnist/mnist_autoencoder" 17 | momentum: 0.9 18 | # solver mode: CPU or GPU 19 | solver_mode: GPU 20 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/mnist/mnist_autoencoder_solver_adadelta.prototxt: -------------------------------------------------------------------------------- 1 | net: "examples/mnist/mnist_autoencoder.prototxt" 2 | test_state: { stage: 'test-on-train' } 3 | test_iter: 500 4 | test_state: { stage: 'test-on-test' } 5 | test_iter: 100 6 | test_interval: 500 7 | test_compute_loss: true 8 | base_lr: 1.0 9 | lr_policy: "fixed" 10 | momentum: 0.95 11 | delta: 1e-8 12 | display: 100 13 | max_iter: 65000 14 | weight_decay: 0.0005 15 | snapshot: 10000 16 | snapshot_prefix: "examples/mnist/mnist_autoencoder_adadelta_train" 17 | # solver mode: CPU or GPU 18 | solver_mode: GPU 19 | type: "AdaDelta" 20 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/mnist/mnist_autoencoder_solver_adagrad.prototxt: -------------------------------------------------------------------------------- 1 | net: "examples/mnist/mnist_autoencoder.prototxt" 2 | test_state: { stage: 'test-on-train' } 3 | test_iter: 500 4 | test_state: { stage: 'test-on-test' } 5 | test_iter: 100 6 | test_interval: 500 7 | test_compute_loss: true 8 | base_lr: 0.01 9 | lr_policy: "fixed" 10 | display: 100 11 | max_iter: 65000 12 | weight_decay: 0.0005 13 | snapshot: 10000 14 | snapshot_prefix: "examples/mnist/mnist_autoencoder_adagrad_train" 15 | # solver mode: CPU or GPU 16 | solver_mode: GPU 17 | type: "AdaGrad" 18 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/mnist/mnist_autoencoder_solver_nesterov.prototxt: -------------------------------------------------------------------------------- 1 | net: "examples/mnist/mnist_autoencoder.prototxt" 2 | test_state: { stage: 'test-on-train' } 3 | test_iter: 500 4 | test_state: { stage: 'test-on-test' } 5 | test_iter: 100 6 | test_interval: 500 7 | test_compute_loss: true 8 | base_lr: 0.01 9 | lr_policy: "step" 10 | gamma: 0.1 11 | stepsize: 10000 12 | display: 100 13 | max_iter: 65000 14 | weight_decay: 0.0005 15 | snapshot: 10000 16 | snapshot_prefix: "examples/mnist/mnist_autoencoder_nesterov_train" 17 | momentum: 0.95 18 | # solver mode: CPU or GPU 19 | solver_mode: GPU 20 | type: "Nesterov" 21 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/mnist/train_lenet.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env sh 2 | 3 | ./build/tools/caffe train --solver=examples/mnist/lenet_solver.prototxt 4 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/mnist/train_lenet_adam.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env sh 2 | 3 | ./build/tools/caffe train --solver=examples/mnist/lenet_solver_adam.prototxt 4 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/mnist/train_lenet_consolidated.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env sh 2 | 3 | ./build/tools/caffe train \ 4 | --solver=examples/mnist/lenet_consolidated_solver.prototxt 5 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/mnist/train_lenet_rmsprop.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env sh 2 | 3 | ./build/tools/caffe train --solver=examples/mnist/lenet_solver_rmsprop.prototxt 4 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/mnist/train_mnist_autoencoder.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env sh 2 | 3 | ./build/tools/caffe train \ 4 | --solver=examples/mnist/mnist_autoencoder_solver.prototxt 5 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/mnist/train_mnist_autoencoder_adadelta.sh: -------------------------------------------------------------------------------- 1 | #!/bin/bash 2 | 3 | ./build/tools/caffe train \ 4 | --solver=examples/mnist/mnist_autoencoder_solver_adadelta.prototxt 5 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/mnist/train_mnist_autoencoder_adagrad.sh: -------------------------------------------------------------------------------- 1 | #!/bin/bash 2 | 3 | ./build/tools/caffe train \ 4 | --solver=examples/mnist/mnist_autoencoder_solver_adagrad.prototxt 5 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/mnist/train_mnist_autoencoder_nesterov.sh: -------------------------------------------------------------------------------- 1 | #!/bin/bash 2 | 3 | ./build/tools/caffe train \ 4 | --solver=examples/mnist/mnist_autoencoder_solver_nesterov.prototxt 5 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/net_surgery/conv.prototxt: -------------------------------------------------------------------------------- 1 | # Simple single-layer network to showcase editing model parameters. 2 | name: "convolution" 3 | input: "data" 4 | input_shape { 5 | dim: 1 6 | dim: 1 7 | dim: 100 8 | dim: 100 9 | } 10 | layer { 11 | name: "conv" 12 | type: "Convolution" 13 | bottom: "data" 14 | top: "conv" 15 | convolution_param { 16 | num_output: 3 17 | kernel_size: 5 18 | stride: 1 19 | weight_filler { 20 | type: "gaussian" 21 | std: 0.01 22 | } 23 | bias_filler { 24 | type: "constant" 25 | value: 0 26 | } 27 | } 28 | } 29 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/pycaffe/layers/pyloss.py: -------------------------------------------------------------------------------- 1 | import caffe 2 | import numpy as np 3 | 4 | 5 | class EuclideanLossLayer(caffe.Layer): 6 | """ 7 | Compute the Euclidean Loss in the same manner as the C++ EuclideanLossLayer 8 | to demonstrate the class interface for developing layers in Python. 9 | """ 10 | 11 | def setup(self, bottom, top): 12 | # check input pair 13 | if len(bottom) != 2: 14 | raise Exception("Need two inputs to compute distance.") 15 | 16 | def reshape(self, bottom, top): 17 | # check input dimensions match 18 | if bottom[0].count != bottom[1].count: 19 | raise Exception("Inputs must have the same dimension.") 20 | # difference is shape of inputs 21 | self.diff = np.zeros_like(bottom[0].data, dtype=np.float32) 22 | # loss output is scalar 23 | top[0].reshape(1) 24 | 25 | def forward(self, bottom, top): 26 | self.diff[...] = bottom[0].data - bottom[1].data 27 | top[0].data[...] = np.sum(self.diff**2) / bottom[0].num / 2. 28 | 29 | def backward(self, top, propagate_down, bottom): 30 | for i in range(2): 31 | if not propagate_down[i]: 32 | continue 33 | if i == 0: 34 | sign = 1 35 | else: 36 | sign = -1 37 | bottom[i].diff[...] = sign * self.diff / bottom[i].num 38 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/pycaffe/linreg.prototxt: -------------------------------------------------------------------------------- 1 | name: 'LinearRegressionExample' 2 | # define a simple network for linear regression on dummy data 3 | # that computes the loss by a PythonLayer. 4 | layer { 5 | type: 'DummyData' 6 | name: 'x' 7 | top: 'x' 8 | dummy_data_param { 9 | shape: { dim: 10 dim: 3 dim: 2 } 10 | data_filler: { type: 'gaussian' } 11 | } 12 | } 13 | layer { 14 | type: 'DummyData' 15 | name: 'y' 16 | top: 'y' 17 | dummy_data_param { 18 | shape: { dim: 10 dim: 3 dim: 2 } 19 | data_filler: { type: 'gaussian' } 20 | } 21 | } 22 | # include InnerProduct layers for parameters 23 | # so the net will need backward 24 | layer { 25 | type: 'InnerProduct' 26 | name: 'ipx' 27 | top: 'ipx' 28 | bottom: 'x' 29 | inner_product_param { 30 | num_output: 10 31 | weight_filler { type: 'xavier' } 32 | } 33 | } 34 | layer { 35 | type: 'InnerProduct' 36 | name: 'ipy' 37 | top: 'ipy' 38 | bottom: 'y' 39 | inner_product_param { 40 | num_output: 10 41 | weight_filler { type: 'xavier' } 42 | } 43 | } 44 | layer { 45 | type: 'Python' 46 | name: 'loss' 47 | top: 'loss' 48 | bottom: 'ipx' 49 | bottom: 'ipy' 50 | python_param { 51 | # the module name -- usually the filename -- that needs to be in $PYTHONPATH 52 | module: 'pyloss' 53 | # the layer name -- the class name in the module 54 | layer: 'EuclideanLossLayer' 55 | } 56 | # set loss weight so Caffe knows this is a loss layer. 57 | # since PythonLayer inherits directly from Layer, this isn't automatically 58 | # known to Caffe 59 | loss_weight: 1 60 | } 61 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/siamese/create_mnist_siamese.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env sh 2 | # This script converts the mnist data into leveldb format. 3 | 4 | EXAMPLES=./build/examples/siamese 5 | DATA=./data/mnist 6 | 7 | echo "Creating leveldb..." 8 | 9 | rm -rf ./examples/siamese/mnist_siamese_train_leveldb 10 | rm -rf ./examples/siamese/mnist_siamese_test_leveldb 11 | 12 | $EXAMPLES/convert_mnist_siamese_data.bin \ 13 | $DATA/train-images-idx3-ubyte \ 14 | $DATA/train-labels-idx1-ubyte \ 15 | ./examples/siamese/mnist_siamese_train_leveldb 16 | $EXAMPLES/convert_mnist_siamese_data.bin \ 17 | $DATA/t10k-images-idx3-ubyte \ 18 | $DATA/t10k-labels-idx1-ubyte \ 19 | ./examples/siamese/mnist_siamese_test_leveldb 20 | 21 | echo "Done." 22 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/siamese/mnist_siamese_solver.prototxt: -------------------------------------------------------------------------------- 1 | # The train/test net protocol buffer definition 2 | net: "examples/siamese/mnist_siamese_train_test.prototxt" 3 | # test_iter specifies how many forward passes the test should carry out. 4 | # In the case of MNIST, we have test batch size 100 and 100 test iterations, 5 | # covering the full 10,000 testing images. 6 | test_iter: 100 7 | # Carry out testing every 500 training iterations. 8 | test_interval: 500 9 | # The base learning rate, momentum and the weight decay of the network. 10 | base_lr: 0.01 11 | momentum: 0.9 12 | weight_decay: 0.0000 13 | # The learning rate policy 14 | lr_policy: "inv" 15 | gamma: 0.0001 16 | power: 0.75 17 | # Display every 100 iterations 18 | display: 100 19 | # The maximum number of iterations 20 | max_iter: 50000 21 | # snapshot intermediate results 22 | snapshot: 5000 23 | snapshot_prefix: "examples/siamese/mnist_siamese" 24 | # solver mode: CPU or GPU 25 | solver_mode: GPU 26 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/siamese/train_mnist_siamese.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env sh 2 | 3 | TOOLS=./build/tools 4 | 5 | $TOOLS/caffe train --solver=examples/siamese/mnist_siamese_solver.prototxt 6 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/web_demo/exifutil.py: -------------------------------------------------------------------------------- 1 | """ 2 | This script handles the skimage exif problem. 3 | """ 4 | 5 | from PIL import Image 6 | import numpy as np 7 | 8 | ORIENTATIONS = { # used in apply_orientation 9 | 2: (Image.FLIP_LEFT_RIGHT,), 10 | 3: (Image.ROTATE_180,), 11 | 4: (Image.FLIP_TOP_BOTTOM,), 12 | 5: (Image.FLIP_LEFT_RIGHT, Image.ROTATE_90), 13 | 6: (Image.ROTATE_270,), 14 | 7: (Image.FLIP_LEFT_RIGHT, Image.ROTATE_270), 15 | 8: (Image.ROTATE_90,) 16 | } 17 | 18 | 19 | def open_oriented_im(im_path): 20 | im = Image.open(im_path) 21 | if hasattr(im, '_getexif'): 22 | exif = im._getexif() 23 | if exif is not None and 274 in exif: 24 | orientation = exif[274] 25 | im = apply_orientation(im, orientation) 26 | img = np.asarray(im).astype(np.float32) / 255. 27 | if img.ndim == 2: 28 | img = img[:, :, np.newaxis] 29 | img = np.tile(img, (1, 1, 3)) 30 | elif img.shape[2] == 4: 31 | img = img[:, :, :3] 32 | return img 33 | 34 | 35 | def apply_orientation(im, orientation): 36 | if orientation in ORIENTATIONS: 37 | for method in ORIENTATIONS[orientation]: 38 | im = im.transpose(method) 39 | return im 40 | -------------------------------------------------------------------------------- /deeplab-caffe/examples/web_demo/requirements.txt: -------------------------------------------------------------------------------- 1 | werkzeug 2 | flask 3 | tornado 4 | numpy 5 | pandas 6 | pillow 7 | pyyaml 8 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/caffe.hpp: -------------------------------------------------------------------------------- 1 | // caffe.hpp is the header file that you need to include in your code. It wraps 2 | // all the internal caffe header files into one for simpler inclusion. 3 | 4 | #ifndef CAFFE_CAFFE_HPP_ 5 | #define CAFFE_CAFFE_HPP_ 6 | 7 | #include "caffe/blob.hpp" 8 | #include "caffe/common.hpp" 9 | #include "caffe/filler.hpp" 10 | #include "caffe/layer.hpp" 11 | #include "caffe/layer_factory.hpp" 12 | #include "caffe/net.hpp" 13 | #include "caffe/parallel.hpp" 14 | #include "caffe/proto/caffe.pb.h" 15 | #include "caffe/solver.hpp" 16 | #include "caffe/solver_factory.hpp" 17 | #include "caffe/util/benchmark.hpp" 18 | #include "caffe/util/io.hpp" 19 | #include "caffe/util/upgrade_proto.hpp" 20 | 21 | #endif // CAFFE_CAFFE_HPP_ 22 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/common.cuh: -------------------------------------------------------------------------------- 1 | // Copyright 2014 George Papandreou 2 | 3 | #ifndef CAFFE_COMMON_CUH_ 4 | #define CAFFE_COMMON_CUH_ 5 | 6 | #include 7 | 8 | // CUDA: atomicAdd is not defined for doubles 9 | #if !defined(__CUDA_ARCH__) || __CUDA_ARCH__ >= 600 10 | #else 11 | static __inline__ __device__ double atomicAdd(double *address, double val) { 12 | unsigned long long int* address_as_ull = (unsigned long long int*)address; 13 | unsigned long long int old = *address_as_ull, assumed; 14 | if (val==0.0) 15 | return __longlong_as_double(old); 16 | do { 17 | assumed = old; 18 | old = atomicCAS(address_as_ull, assumed, __double_as_longlong(val +__longlong_as_double(assumed))); 19 | } while (assumed != old); 20 | return __longlong_as_double(old); 21 | } 22 | #endif 23 | 24 | #endif 25 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/internal_thread.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_INTERNAL_THREAD_HPP_ 2 | #define CAFFE_INTERNAL_THREAD_HPP_ 3 | 4 | #include "caffe/common.hpp" 5 | 6 | /** 7 | Forward declare boost::thread instead of including boost/thread.hpp 8 | to avoid a boost/NVCC issues (#1009, #1010) on OSX. 9 | */ 10 | namespace boost { class thread; } 11 | 12 | namespace caffe { 13 | 14 | /** 15 | * Virtual class encapsulate boost::thread for use in base class 16 | * The child class will acquire the ability to run a single thread, 17 | * by reimplementing the virtual function InternalThreadEntry. 18 | */ 19 | class InternalThread { 20 | public: 21 | InternalThread() : thread_() {} 22 | virtual ~InternalThread(); 23 | 24 | /** 25 | * Caffe's thread local state will be initialized using the current 26 | * thread values, e.g. device id, solver index etc. The random seed 27 | * is initialized using caffe_rng_rand. 28 | */ 29 | void StartInternalThread(); 30 | 31 | /** Will not return until the internal thread has exited. */ 32 | void StopInternalThread(); 33 | 34 | bool is_started() const; 35 | 36 | protected: 37 | /* Implement this method in your subclass 38 | with the code you want your thread to run. */ 39 | virtual void InternalThreadEntry() {} 40 | 41 | /* Should be tested when running loops to exit when requested. */ 42 | bool must_stop(); 43 | 44 | private: 45 | void entry(int device, Caffe::Brew mode, int rand_seed, int solver_count, 46 | bool root_solver); 47 | 48 | shared_ptr thread_; 49 | }; 50 | 51 | } // namespace caffe 52 | 53 | #endif // CAFFE_INTERNAL_THREAD_HPP_ 54 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/layers/cudnn_lcn_layer.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_CUDNN_LCN_LAYER_HPP_ 2 | #define CAFFE_CUDNN_LCN_LAYER_HPP_ 3 | 4 | #include 5 | 6 | #include "caffe/blob.hpp" 7 | #include "caffe/layer.hpp" 8 | #include "caffe/proto/caffe.pb.h" 9 | 10 | #include "caffe/layers/lrn_layer.hpp" 11 | #include "caffe/layers/power_layer.hpp" 12 | 13 | namespace caffe { 14 | 15 | #ifdef USE_CUDNN 16 | template 17 | class CuDNNLCNLayer : public LRNLayer { 18 | public: 19 | explicit CuDNNLCNLayer(const LayerParameter& param) 20 | : LRNLayer(param), handles_setup_(false), tempDataSize(0), 21 | tempData1(NULL), tempData2(NULL) {} 22 | virtual void LayerSetUp(const vector*>& bottom, 23 | const vector*>& top); 24 | virtual void Reshape(const vector*>& bottom, 25 | const vector*>& top); 26 | virtual ~CuDNNLCNLayer(); 27 | 28 | protected: 29 | virtual void Forward_gpu(const vector*>& bottom, 30 | const vector*>& top); 31 | virtual void Backward_gpu(const vector*>& top, 32 | const vector& propagate_down, const vector*>& bottom); 33 | 34 | bool handles_setup_; 35 | cudnnHandle_t handle_; 36 | cudnnLRNDescriptor_t norm_desc_; 37 | cudnnTensorDescriptor_t bottom_desc_, top_desc_; 38 | 39 | int size_, pre_pad_; 40 | Dtype alpha_, beta_, k_; 41 | 42 | size_t tempDataSize; 43 | void *tempData1, *tempData2; 44 | }; 45 | #endif 46 | 47 | } // namespace caffe 48 | 49 | #endif // CAFFE_CUDNN_LCN_LAYER_HPP_ 50 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/layers/cudnn_lrn_layer.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_CUDNN_LRN_LAYER_HPP_ 2 | #define CAFFE_CUDNN_LRN_LAYER_HPP_ 3 | 4 | #include 5 | 6 | #include "caffe/blob.hpp" 7 | #include "caffe/layer.hpp" 8 | #include "caffe/proto/caffe.pb.h" 9 | 10 | #include "caffe/layers/lrn_layer.hpp" 11 | 12 | namespace caffe { 13 | 14 | #ifdef USE_CUDNN 15 | template 16 | class CuDNNLRNLayer : public LRNLayer { 17 | public: 18 | explicit CuDNNLRNLayer(const LayerParameter& param) 19 | : LRNLayer(param), handles_setup_(false) {} 20 | virtual void LayerSetUp(const vector*>& bottom, 21 | const vector*>& top); 22 | virtual void Reshape(const vector*>& bottom, 23 | const vector*>& top); 24 | virtual ~CuDNNLRNLayer(); 25 | 26 | protected: 27 | virtual void Forward_gpu(const vector*>& bottom, 28 | const vector*>& top); 29 | virtual void Backward_gpu(const vector*>& top, 30 | const vector& propagate_down, const vector*>& bottom); 31 | 32 | bool handles_setup_; 33 | cudnnHandle_t handle_; 34 | cudnnLRNDescriptor_t norm_desc_; 35 | cudnnTensorDescriptor_t bottom_desc_, top_desc_; 36 | 37 | int size_; 38 | Dtype alpha_, beta_, k_; 39 | }; 40 | #endif 41 | 42 | } // namespace caffe 43 | 44 | #endif // CAFFE_CUDNN_LRN_LAYER_HPP_ 45 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/layers/cudnn_relu_layer.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_CUDNN_RELU_LAYER_HPP_ 2 | #define CAFFE_CUDNN_RELU_LAYER_HPP_ 3 | 4 | #include 5 | 6 | #include "caffe/blob.hpp" 7 | #include "caffe/layer.hpp" 8 | #include "caffe/proto/caffe.pb.h" 9 | 10 | #include "caffe/layers/neuron_layer.hpp" 11 | #include "caffe/layers/relu_layer.hpp" 12 | 13 | namespace caffe { 14 | 15 | #ifdef USE_CUDNN 16 | /** 17 | * @brief CuDNN acceleration of ReLULayer. 18 | */ 19 | template 20 | class CuDNNReLULayer : public ReLULayer { 21 | public: 22 | explicit CuDNNReLULayer(const LayerParameter& param) 23 | : ReLULayer(param), handles_setup_(false) {} 24 | virtual void LayerSetUp(const vector*>& bottom, 25 | const vector*>& top); 26 | virtual void Reshape(const vector*>& bottom, 27 | const vector*>& top); 28 | virtual ~CuDNNReLULayer(); 29 | 30 | protected: 31 | virtual void Forward_gpu(const vector*>& bottom, 32 | const vector*>& top); 33 | virtual void Backward_gpu(const vector*>& top, 34 | const vector& propagate_down, const vector*>& bottom); 35 | 36 | bool handles_setup_; 37 | cudnnHandle_t handle_; 38 | cudnnTensorDescriptor_t bottom_desc_; 39 | cudnnTensorDescriptor_t top_desc_; 40 | }; 41 | #endif 42 | 43 | } // namespace caffe 44 | 45 | #endif // CAFFE_CUDNN_RELU_LAYER_HPP_ 46 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/layers/cudnn_sigmoid_layer.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_CUDNN_SIGMOID_LAYER_HPP_ 2 | #define CAFFE_CUDNN_SIGMOID_LAYER_HPP_ 3 | 4 | #include 5 | 6 | #include "caffe/blob.hpp" 7 | #include "caffe/layer.hpp" 8 | #include "caffe/proto/caffe.pb.h" 9 | 10 | #include "caffe/layers/neuron_layer.hpp" 11 | #include "caffe/layers/sigmoid_layer.hpp" 12 | 13 | namespace caffe { 14 | 15 | #ifdef USE_CUDNN 16 | /** 17 | * @brief CuDNN acceleration of SigmoidLayer. 18 | */ 19 | template 20 | class CuDNNSigmoidLayer : public SigmoidLayer { 21 | public: 22 | explicit CuDNNSigmoidLayer(const LayerParameter& param) 23 | : SigmoidLayer(param), handles_setup_(false) {} 24 | virtual void LayerSetUp(const vector*>& bottom, 25 | const vector*>& top); 26 | virtual void Reshape(const vector*>& bottom, 27 | const vector*>& top); 28 | virtual ~CuDNNSigmoidLayer(); 29 | 30 | protected: 31 | virtual void Forward_gpu(const vector*>& bottom, 32 | const vector*>& top); 33 | virtual void Backward_gpu(const vector*>& top, 34 | const vector& propagate_down, const vector*>& bottom); 35 | 36 | bool handles_setup_; 37 | cudnnHandle_t handle_; 38 | cudnnTensorDescriptor_t bottom_desc_; 39 | cudnnTensorDescriptor_t top_desc_; 40 | }; 41 | #endif 42 | 43 | } // namespace caffe 44 | 45 | #endif // CAFFE_CUDNN_SIGMOID_LAYER_HPP_ 46 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/layers/cudnn_softmax_layer.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_CUDNN_SOFTMAX_LAYER_HPP_ 2 | #define CAFFE_CUDNN_SOFTMAX_LAYER_HPP_ 3 | 4 | #include 5 | 6 | #include "caffe/blob.hpp" 7 | #include "caffe/layer.hpp" 8 | #include "caffe/proto/caffe.pb.h" 9 | 10 | #include "caffe/layers/softmax_layer.hpp" 11 | 12 | namespace caffe { 13 | 14 | #ifdef USE_CUDNN 15 | /** 16 | * @brief cuDNN implementation of SoftmaxLayer. 17 | * Fallback to SoftmaxLayer for CPU mode. 18 | */ 19 | template 20 | class CuDNNSoftmaxLayer : public SoftmaxLayer { 21 | public: 22 | explicit CuDNNSoftmaxLayer(const LayerParameter& param) 23 | : SoftmaxLayer(param), handles_setup_(false) {} 24 | virtual void LayerSetUp(const vector*>& bottom, 25 | const vector*>& top); 26 | virtual void Reshape(const vector*>& bottom, 27 | const vector*>& top); 28 | virtual ~CuDNNSoftmaxLayer(); 29 | 30 | protected: 31 | virtual void Forward_gpu(const vector*>& bottom, 32 | const vector*>& top); 33 | virtual void Backward_gpu(const vector*>& top, 34 | const vector& propagate_down, const vector*>& bottom); 35 | 36 | bool handles_setup_; 37 | cudnnHandle_t handle_; 38 | cudnnTensorDescriptor_t bottom_desc_; 39 | cudnnTensorDescriptor_t top_desc_; 40 | }; 41 | #endif 42 | 43 | } // namespace caffe 44 | 45 | #endif // CAFFE_CUDNN_SOFTMAX_LAYER_HPP_ 46 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/layers/cudnn_tanh_layer.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_CUDNN_TANH_LAYER_HPP_ 2 | #define CAFFE_CUDNN_TANH_LAYER_HPP_ 3 | 4 | #include 5 | 6 | #include "caffe/blob.hpp" 7 | #include "caffe/layer.hpp" 8 | #include "caffe/proto/caffe.pb.h" 9 | 10 | #include "caffe/layers/neuron_layer.hpp" 11 | #include "caffe/layers/tanh_layer.hpp" 12 | 13 | namespace caffe { 14 | 15 | #ifdef USE_CUDNN 16 | /** 17 | * @brief CuDNN acceleration of TanHLayer. 18 | */ 19 | template 20 | class CuDNNTanHLayer : public TanHLayer { 21 | public: 22 | explicit CuDNNTanHLayer(const LayerParameter& param) 23 | : TanHLayer(param), handles_setup_(false) {} 24 | virtual void LayerSetUp(const vector*>& bottom, 25 | const vector*>& top); 26 | virtual void Reshape(const vector*>& bottom, 27 | const vector*>& top); 28 | virtual ~CuDNNTanHLayer(); 29 | 30 | protected: 31 | virtual void Forward_gpu(const vector*>& bottom, 32 | const vector*>& top); 33 | virtual void Backward_gpu(const vector*>& top, 34 | const vector& propagate_down, const vector*>& bottom); 35 | 36 | bool handles_setup_; 37 | cudnnHandle_t handle_; 38 | cudnnTensorDescriptor_t bottom_desc_; 39 | cudnnTensorDescriptor_t top_desc_; 40 | }; 41 | #endif 42 | 43 | } // namespace caffe 44 | 45 | #endif // CAFFE_CUDNN_TANH_LAYER_HPP_ 46 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/layers/data_layer.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_DATA_LAYER_HPP_ 2 | #define CAFFE_DATA_LAYER_HPP_ 3 | 4 | #include 5 | 6 | #include "caffe/blob.hpp" 7 | #include "caffe/data_reader.hpp" 8 | #include "caffe/data_transformer.hpp" 9 | #include "caffe/internal_thread.hpp" 10 | #include "caffe/layer.hpp" 11 | #include "caffe/layers/base_data_layer.hpp" 12 | #include "caffe/proto/caffe.pb.h" 13 | #include "caffe/util/db.hpp" 14 | 15 | namespace caffe { 16 | 17 | template 18 | class DataLayer : public BasePrefetchingDataLayer { 19 | public: 20 | explicit DataLayer(const LayerParameter& param); 21 | virtual ~DataLayer(); 22 | virtual void DataLayerSetUp(const vector*>& bottom, 23 | const vector*>& top); 24 | // DataLayer uses DataReader instead for sharing for parallelism 25 | virtual inline bool ShareInParallel() const { return false; } 26 | virtual inline const char* type() const { return "Data"; } 27 | virtual inline int ExactNumBottomBlobs() const { return 0; } 28 | virtual inline int MinTopBlobs() const { return 1; } 29 | virtual inline int MaxTopBlobs() const { return 2; } 30 | 31 | protected: 32 | virtual void load_batch(Batch* batch); 33 | 34 | DataReader reader_; 35 | }; 36 | 37 | } // namespace caffe 38 | 39 | #endif // CAFFE_DATA_LAYER_HPP_ 40 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/layers/get_data_dim_layer.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_GET_DATA_DIM_LAYER_HPP_ 2 | #define CAFFE_GET_DATA_DIM_LAYER_HPP_ 3 | 4 | #include 5 | #include 6 | 7 | #include "caffe/blob.hpp" 8 | #include "caffe/layer.hpp" 9 | #include "caffe/proto/caffe.pb.h" 10 | 11 | namespace caffe { 12 | 13 | /* 14 | GetDataDimLayer 15 | */ 16 | template 17 | class GetDataDimLayer : public Layer { 18 | public: 19 | explicit GetDataDimLayer(const LayerParameter& param) 20 | : Layer(param) {} 21 | virtual void LayerSetUp(const vector*>& bottom, 22 | const vector*>& top); 23 | virtual void Reshape(const vector*>& bottom, 24 | const vector*>& top); 25 | virtual inline const char* type() const { return "GetDataDim"; } 26 | virtual inline int ExactNumTopBlobs() const { return 1; } 27 | 28 | protected: 29 | virtual void Forward_cpu(const vector*>& bottom, 30 | const vector*>& top); 31 | virtual void Backward_cpu(const vector*>& top, 32 | const vector& propagate_down, const vector*>& bottom); 33 | }; 34 | 35 | } // namespace caffe 36 | 37 | #endif // CAFFE_GET_DATA_DIM_LAYER_HPP_ -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/layers/image_cls_data_layer.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_IMAGE_CLS_DATA_LAYER_HPP_ 2 | #define CAFFE_IMAGE_CLS_DATA_LAYER_HPP_ 3 | 4 | #include 5 | #include 6 | #include 7 | 8 | #include "caffe/blob.hpp" 9 | #include "caffe/data_transformer.hpp" 10 | #include "caffe/internal_thread.hpp" 11 | #include "caffe/layer.hpp" 12 | #include "caffe/layers/base_data_layer.hpp" 13 | #include "caffe/proto/caffe.pb.h" 14 | 15 | 16 | namespace caffe { 17 | 18 | template 19 | class ImageClsDataLayer : public ImageDimPrefetchingDataLayer { 20 | public: 21 | explicit ImageClsDataLayer(const LayerParameter& param) 22 | : ImageDimPrefetchingDataLayer(param) {} 23 | virtual ~ImageClsDataLayer(); 24 | virtual void DataLayerSetUp(const vector*>& bottom, 25 | const vector*>& top); 26 | 27 | virtual inline const char* type() const { return "ImageClsData"; } 28 | virtual inline int ExactNumBottomBlobs() const { return 0; } 29 | virtual inline int ExactNumTopBlobs() const { return 3; } 30 | virtual inline bool AutoTopBlobs() const { return true; } 31 | 32 | protected: 33 | virtual void ShuffleImages(); 34 | virtual void load_batch(Batch* batch); 35 | 36 | Blob transformed_label_; 37 | shared_ptr prefetch_rng_; 38 | vector > lines_; 39 | int lines_id_; 40 | }; 41 | 42 | } // namespace caffe 43 | 44 | #endif // CAFFE_IMAGE_CLS_DATA_LAYER_HPP_ 45 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/layers/image_data_layer.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_IMAGE_DATA_LAYER_HPP_ 2 | #define CAFFE_IMAGE_DATA_LAYER_HPP_ 3 | 4 | #include 5 | #include 6 | #include 7 | 8 | #include "caffe/blob.hpp" 9 | #include "caffe/data_transformer.hpp" 10 | #include "caffe/internal_thread.hpp" 11 | #include "caffe/layer.hpp" 12 | #include "caffe/layers/base_data_layer.hpp" 13 | #include "caffe/proto/caffe.pb.h" 14 | 15 | namespace caffe { 16 | 17 | /** 18 | * @brief Provides data to the Net from image files. 19 | * 20 | * TODO(dox): thorough documentation for Forward and proto params. 21 | */ 22 | template 23 | class ImageDataLayer : public BasePrefetchingDataLayer { 24 | public: 25 | explicit ImageDataLayer(const LayerParameter& param) 26 | : BasePrefetchingDataLayer(param) {} 27 | virtual ~ImageDataLayer(); 28 | virtual void DataLayerSetUp(const vector*>& bottom, 29 | const vector*>& top); 30 | 31 | virtual inline const char* type() const { return "ImageData"; } 32 | virtual inline int ExactNumBottomBlobs() const { return 0; } 33 | virtual inline int ExactNumTopBlobs() const { return 2; } 34 | 35 | protected: 36 | shared_ptr prefetch_rng_; 37 | virtual void ShuffleImages(); 38 | virtual void load_batch(Batch* batch); 39 | 40 | vector > lines_; 41 | int lines_id_; 42 | }; 43 | 44 | 45 | } // namespace caffe 46 | 47 | #endif // CAFFE_IMAGE_DATA_LAYER_HPP_ 48 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/layers/image_seg_data_layer.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_IMAGE_SEG_DATA_LAYER_HPP_ 2 | #define CAFFE_IMAGE_SEG_DATA_LAYER_HPP_ 3 | 4 | #include 5 | #include 6 | #include 7 | 8 | #include "caffe/blob.hpp" 9 | #include "caffe/data_transformer.hpp" 10 | #include "caffe/internal_thread.hpp" 11 | #include "caffe/layer.hpp" 12 | #include "caffe/layers/base_data_layer.hpp" 13 | #include "caffe/proto/caffe.pb.h" 14 | 15 | 16 | namespace caffe { 17 | 18 | template 19 | class ImageSegDataLayer : public ImageDimPrefetchingDataLayer { 20 | public: 21 | explicit ImageSegDataLayer(const LayerParameter& param) 22 | : ImageDimPrefetchingDataLayer(param) {} 23 | virtual ~ImageSegDataLayer(); 24 | virtual void DataLayerSetUp(const vector*>& bottom, 25 | const vector*>& top); 26 | 27 | virtual inline const char* type() const { return "ImageSegData"; } 28 | virtual inline int ExactNumBottomBlobs() const { return 0; } 29 | virtual inline int ExactNumTopBlobs() const { return 3; } 30 | virtual inline bool AutoTopBlobs() const { return true; } 31 | 32 | protected: 33 | virtual void ShuffleImages(); 34 | virtual void load_batch(Batch* batch); 35 | 36 | Blob transformed_label_; 37 | shared_ptr prefetch_rng_; 38 | vector > lines_; 39 | int lines_id_; 40 | }; 41 | 42 | } // namespace caffe 43 | 44 | #endif // CAFFE_IMAGE_SEG_DATA_LAYER_HPP_ 45 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/layers/mat_read_layer.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_MAT_READ_LAYER_HPP_ 2 | #define CAFFE_MAT_READ_LAYER_HPP_ 3 | 4 | #include 5 | #include 6 | 7 | #include "caffe/blob.hpp" 8 | #include "caffe/layer.hpp" 9 | #include "caffe/proto/caffe.pb.h" 10 | 11 | namespace caffe { 12 | 13 | template 14 | class MatReadLayer : public Layer { 15 | public: 16 | explicit MatReadLayer(const LayerParameter& param) 17 | : Layer(param) {} 18 | virtual void LayerSetUp(const vector*>& bottom, 19 | const vector*>& top); 20 | virtual void Reshape(const vector*>& bottom, 21 | const vector*>& top); 22 | virtual inline const char* type() const { return "MatRead"; } 23 | virtual inline int ExactNumBottomBlobs() const { return 0; } 24 | virtual inline int ExactNumTopBlobs() const { return 1; } 25 | 26 | protected: 27 | virtual void Forward_cpu(const vector*>& bottom, 28 | const vector*>& top); 29 | virtual void Backward_cpu(const vector*>& top, 30 | const vector& propagate_down, const vector*>& bottom); 31 | 32 | bool reshape_; 33 | int batch_size_; 34 | int iter_; 35 | string prefix_; 36 | vector fnames_; 37 | }; 38 | 39 | } // namespace caffe 40 | 41 | #endif // CAFFE_MAT_READ_LAYER_HPP_ 42 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/layers/mat_write_layer.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_MAT_WRITE_LAYER_HPP_ 2 | #define CAFFE_MAT_WRITE_LAYER_HPP_ 3 | 4 | #include 5 | #include 6 | 7 | #include "caffe/blob.hpp" 8 | #include "caffe/layer.hpp" 9 | #include "caffe/proto/caffe.pb.h" 10 | 11 | namespace caffe { 12 | 13 | /* 14 | MatWriteLayer 15 | */ 16 | template 17 | class MatWriteLayer : public Layer { 18 | public: 19 | explicit MatWriteLayer(const LayerParameter& param) 20 | : Layer(param) {} 21 | virtual void LayerSetUp(const vector*>& bottom, 22 | const vector*>& top); 23 | virtual void Reshape(const vector*>& bottom, 24 | const vector*>& top); 25 | virtual inline const char* type() const { return "MatWrite"; } 26 | virtual inline int ExactNumTopBlobs() const { return 0; } 27 | 28 | protected: 29 | virtual void Forward_cpu(const vector*>& bottom, 30 | const vector*>& top); 31 | virtual void Backward_cpu(const vector*>& top, 32 | const vector& propagate_down, const vector*>& bottom); 33 | 34 | int iter_; 35 | int period_; 36 | string prefix_; 37 | vector fnames_; 38 | }; 39 | 40 | } // namespace caffe 41 | 42 | #endif // CAFFE_MAT_WRITE_LAYER_HPP_ 43 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/layers/neuron_layer.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_NEURON_LAYER_HPP_ 2 | #define CAFFE_NEURON_LAYER_HPP_ 3 | 4 | #include 5 | 6 | #include "caffe/blob.hpp" 7 | #include "caffe/layer.hpp" 8 | #include "caffe/proto/caffe.pb.h" 9 | 10 | namespace caffe { 11 | 12 | /** 13 | * @brief An interface for layers that take one blob as input (@f$ x @f$) 14 | * and produce one equally-sized blob as output (@f$ y @f$), where 15 | * each element of the output depends only on the corresponding input 16 | * element. 17 | */ 18 | template 19 | class NeuronLayer : public Layer { 20 | public: 21 | explicit NeuronLayer(const LayerParameter& param) 22 | : Layer(param) {} 23 | virtual void Reshape(const vector*>& bottom, 24 | const vector*>& top); 25 | 26 | virtual inline int ExactNumBottomBlobs() const { return 1; } 27 | virtual inline int ExactNumTopBlobs() const { return 1; } 28 | }; 29 | 30 | } // namespace caffe 31 | 32 | #endif // CAFFE_NEURON_LAYER_HPP_ 33 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/layers/tile_layer.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_TILE_LAYER_HPP_ 2 | #define CAFFE_TILE_LAYER_HPP_ 3 | 4 | #include 5 | 6 | #include "caffe/blob.hpp" 7 | #include "caffe/layer.hpp" 8 | #include "caffe/proto/caffe.pb.h" 9 | 10 | namespace caffe { 11 | 12 | /** 13 | * @brief Copy a Blob along specified dimensions. 14 | */ 15 | template 16 | class TileLayer : public Layer { 17 | public: 18 | explicit TileLayer(const LayerParameter& param) 19 | : Layer(param) {} 20 | virtual void Reshape(const vector*>& bottom, 21 | const vector*>& top); 22 | 23 | virtual inline const char* type() const { return "Tile"; } 24 | virtual inline int ExactNumBottomBlobs() const { return 1; } 25 | virtual inline int ExactNumTopBlobs() const { return 1; } 26 | 27 | protected: 28 | virtual void Forward_cpu(const vector*>& bottom, 29 | const vector*>& top); 30 | virtual void Forward_gpu(const vector*>& bottom, 31 | const vector*>& top); 32 | 33 | virtual void Backward_cpu(const vector*>& top, 34 | const vector& propagate_down, const vector*>& bottom); 35 | virtual void Backward_gpu(const vector*>& top, 36 | const vector& propagate_down, const vector*>& bottom); 37 | 38 | unsigned int axis_, tiles_, outer_dim_, inner_dim_; 39 | }; 40 | 41 | } // namespace caffe 42 | 43 | #endif // CAFFE_TILE_LAYER_HPP_ 44 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/layers/visual_saliency_layer.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_VISUAL_SALIENCY_LAYER_HPP_ 2 | #define CAFFE_VISUAL_SALIENCY_LAYER_HPP_ 3 | 4 | #include 5 | #include 6 | 7 | #include "caffe/blob.hpp" 8 | #include "caffe/layer.hpp" 9 | #include "caffe/proto/caffe.pb.h" 10 | 11 | namespace caffe { 12 | 13 | /* 14 | VisualSaliencyLayer 15 | */ 16 | template 17 | class VisualSaliencyLayer : public Layer { 18 | public: 19 | explicit VisualSaliencyLayer(const LayerParameter& param) 20 | : Layer(param) {} 21 | virtual void LayerSetUp(const vector*>& bottom, 22 | const vector*>& top); 23 | virtual void Reshape(const vector*>& bottom, 24 | const vector*>& top); 25 | virtual inline const char* type() const { return "VisualSaliency"; } 26 | virtual inline int ExactNumTopBlobs() const { return 0; } 27 | 28 | protected: 29 | virtual void Forward_cpu(const vector*>& bottom, 30 | const vector*>& top); 31 | virtual void Backward_cpu(const vector*>& top, 32 | const vector& propagate_down, const vector*>& bottom); 33 | int iter_; 34 | int visual_interval_; 35 | int visual_num_; 36 | string layer_name_; 37 | string prefix_; 38 | 39 | }; 40 | 41 | } // namespace caffe 42 | 43 | #endif // CAFFE_VISUAL_SALIENCY_LAYER_HPP_ -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/util/benchmark.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_UTIL_BENCHMARK_H_ 2 | #define CAFFE_UTIL_BENCHMARK_H_ 3 | 4 | #include 5 | 6 | #include "caffe/util/device_alternate.hpp" 7 | 8 | namespace caffe { 9 | 10 | class Timer { 11 | public: 12 | Timer(); 13 | virtual ~Timer(); 14 | virtual void Start(); 15 | virtual void Stop(); 16 | virtual float MilliSeconds(); 17 | virtual float MicroSeconds(); 18 | virtual float Seconds(); 19 | 20 | inline bool initted() { return initted_; } 21 | inline bool running() { return running_; } 22 | inline bool has_run_at_least_once() { return has_run_at_least_once_; } 23 | 24 | protected: 25 | void Init(); 26 | 27 | bool initted_; 28 | bool running_; 29 | bool has_run_at_least_once_; 30 | #ifndef CPU_ONLY 31 | cudaEvent_t start_gpu_; 32 | cudaEvent_t stop_gpu_; 33 | #endif 34 | boost::posix_time::ptime start_cpu_; 35 | boost::posix_time::ptime stop_cpu_; 36 | float elapsed_milliseconds_; 37 | float elapsed_microseconds_; 38 | }; 39 | 40 | class CPUTimer : public Timer { 41 | public: 42 | explicit CPUTimer(); 43 | virtual ~CPUTimer() {} 44 | virtual void Start(); 45 | virtual void Stop(); 46 | virtual float MilliSeconds(); 47 | virtual float MicroSeconds(); 48 | }; 49 | 50 | } // namespace caffe 51 | 52 | #endif // CAFFE_UTIL_BENCHMARK_H_ 53 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/util/blocking_queue.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_UTIL_BLOCKING_QUEUE_HPP_ 2 | #define CAFFE_UTIL_BLOCKING_QUEUE_HPP_ 3 | 4 | #include 5 | #include 6 | 7 | namespace caffe { 8 | 9 | template 10 | class BlockingQueue { 11 | public: 12 | explicit BlockingQueue(); 13 | 14 | void push(const T& t); 15 | 16 | bool try_pop(T* t); 17 | 18 | // This logs a message if the threads needs to be blocked 19 | // useful for detecting e.g. when data feeding is too slow 20 | T pop(const string& log_on_wait = ""); 21 | 22 | bool try_peek(T* t); 23 | 24 | // Return element without removing it 25 | T peek(); 26 | 27 | size_t size() const; 28 | 29 | protected: 30 | /** 31 | Move synchronization fields out instead of including boost/thread.hpp 32 | to avoid a boost/NVCC issues (#1009, #1010) on OSX. Also fails on 33 | Linux CUDA 7.0.18. 34 | */ 35 | class sync; 36 | 37 | std::queue queue_; 38 | shared_ptr sync_; 39 | 40 | DISABLE_COPY_AND_ASSIGN(BlockingQueue); 41 | }; 42 | 43 | } // namespace caffe 44 | 45 | #endif 46 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/util/db.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_UTIL_DB_HPP 2 | #define CAFFE_UTIL_DB_HPP 3 | 4 | #include 5 | 6 | #include "caffe/common.hpp" 7 | #include "caffe/proto/caffe.pb.h" 8 | 9 | namespace caffe { namespace db { 10 | 11 | enum Mode { READ, WRITE, NEW }; 12 | 13 | class Cursor { 14 | public: 15 | Cursor() { } 16 | virtual ~Cursor() { } 17 | virtual void SeekToFirst() = 0; 18 | virtual void Next() = 0; 19 | virtual string key() = 0; 20 | virtual string value() = 0; 21 | virtual bool valid() = 0; 22 | 23 | DISABLE_COPY_AND_ASSIGN(Cursor); 24 | }; 25 | 26 | class Transaction { 27 | public: 28 | Transaction() { } 29 | virtual ~Transaction() { } 30 | virtual void Put(const string& key, const string& value) = 0; 31 | virtual void Commit() = 0; 32 | 33 | DISABLE_COPY_AND_ASSIGN(Transaction); 34 | }; 35 | 36 | class DB { 37 | public: 38 | DB() { } 39 | virtual ~DB() { } 40 | virtual void Open(const string& source, Mode mode) = 0; 41 | virtual void Close() = 0; 42 | virtual Cursor* NewCursor() = 0; 43 | virtual Transaction* NewTransaction() = 0; 44 | 45 | DISABLE_COPY_AND_ASSIGN(DB); 46 | }; 47 | 48 | DB* GetDB(DataParameter::DB backend); 49 | DB* GetDB(const string& backend); 50 | 51 | } // namespace db 52 | } // namespace caffe 53 | 54 | #endif // CAFFE_UTIL_DB_HPP 55 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/util/densecrf_pairwise.hpp: -------------------------------------------------------------------------------- 1 | #ifndef _DENSECRF_PAIRWISE_H 2 | #define _DENSECRF_PAIRWISE_H 3 | 4 | #include 5 | #include 6 | 7 | #include "caffe/util/permutohedral.hpp" 8 | 9 | class PairwisePotential { 10 | public: 11 | virtual ~PairwisePotential(); 12 | virtual void apply(float * out_values, const float * in_values, float * tmp, int value_size) const = 0; 13 | }; 14 | 15 | class SemiMetricFunction { 16 | public: 17 | virtual ~SemiMetricFunction(); 18 | // For two probabilities apply 19 | // the semi metric transform: v_i = sum_j mu_ij u_j 20 | virtual void apply(float * out_values, const float * in_values, int value_size) const = 0; 21 | }; 22 | 23 | class PottsPotential: public PairwisePotential{ 24 | protected: 25 | Permutohedral lattice_; 26 | PottsPotential( const PottsPotential& ){} 27 | int N_; 28 | float w_; 29 | float *norm_; 30 | public: 31 | virtual ~PottsPotential(); 32 | PottsPotential(const float* features, int D, int N, float w, bool per_pixel_normalization=true); 33 | 34 | virtual void apply(float* out_values, const float* in_values, float* tmp, int value_size) const; 35 | }; 36 | 37 | class SemiMetricPotential: public PottsPotential{ 38 | protected: 39 | const SemiMetricFunction * function_; 40 | public: 41 | virtual ~SemiMetricPotential(); 42 | virtual void apply(float* out_values, const float* in_values, float* tmp, int value_size) const; 43 | SemiMetricPotential(const float* features, int D, int N, float w, const SemiMetricFunction* function, bool per_pixel_normalization=true); 44 | }; 45 | 46 | 47 | 48 | #endif 49 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/util/format.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_UTIL_FORMAT_H_ 2 | #define CAFFE_UTIL_FORMAT_H_ 3 | 4 | #include // NOLINT(readability/streams) 5 | #include // NOLINT(readability/streams) 6 | #include 7 | 8 | namespace caffe { 9 | 10 | inline std::string format_int(int n, int numberOfLeadingZeros = 0 ) { 11 | std::ostringstream s; 12 | s << std::setw(numberOfLeadingZeros) << std::setfill('0') << n; 13 | return s.str(); 14 | } 15 | 16 | } 17 | 18 | #endif // CAFFE_UTIL_FORMAT_H_ 19 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/util/gpu_util.cuh: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_UTIL_GPU_UTIL_H_ 2 | #define CAFFE_UTIL_GPU_UTIL_H_ 3 | 4 | namespace caffe { 5 | 6 | template 7 | inline __device__ Dtype caffe_gpu_atomic_add(const Dtype val, Dtype* address); 8 | 9 | template <> 10 | inline __device__ 11 | float caffe_gpu_atomic_add(const float val, float* address) { 12 | return atomicAdd(address, val); 13 | } 14 | 15 | // double atomicAdd implementation taken from: 16 | // http://docs.nvidia.com/cuda/cuda-c-programming-guide/#axzz3PVCpVsEG 17 | template <> 18 | inline __device__ 19 | double caffe_gpu_atomic_add(const double val, double* address) { 20 | unsigned long long int* address_as_ull = // NOLINT(runtime/int) 21 | // NOLINT_NEXT_LINE(runtime/int) 22 | reinterpret_cast(address); 23 | unsigned long long int old = *address_as_ull; // NOLINT(runtime/int) 24 | unsigned long long int assumed; // NOLINT(runtime/int) 25 | do { 26 | assumed = old; 27 | old = atomicCAS(address_as_ull, assumed, 28 | __double_as_longlong(val + __longlong_as_double(assumed))); 29 | } while (assumed != old); 30 | return __longlong_as_double(old); 31 | } 32 | 33 | } // namespace caffe 34 | 35 | #endif // CAFFE_UTIL_GPU_UTIL_H_ 36 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/util/hdf5.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_UTIL_HDF5_H_ 2 | #define CAFFE_UTIL_HDF5_H_ 3 | 4 | #include 5 | 6 | #include "hdf5.h" 7 | #include "hdf5_hl.h" 8 | 9 | #include "caffe/blob.hpp" 10 | 11 | namespace caffe { 12 | 13 | template 14 | void hdf5_load_nd_dataset_helper( 15 | hid_t file_id, const char* dataset_name_, int min_dim, int max_dim, 16 | Blob* blob); 17 | 18 | template 19 | void hdf5_load_nd_dataset( 20 | hid_t file_id, const char* dataset_name_, int min_dim, int max_dim, 21 | Blob* blob); 22 | 23 | template 24 | void hdf5_save_nd_dataset( 25 | const hid_t file_id, const string& dataset_name, const Blob& blob, 26 | bool write_diff = false); 27 | 28 | int hdf5_load_int(hid_t loc_id, const string& dataset_name); 29 | void hdf5_save_int(hid_t loc_id, const string& dataset_name, int i); 30 | string hdf5_load_string(hid_t loc_id, const string& dataset_name); 31 | void hdf5_save_string(hid_t loc_id, const string& dataset_name, 32 | const string& s); 33 | 34 | int hdf5_get_num_links(hid_t loc_id); 35 | string hdf5_get_name_by_idx(hid_t loc_id, int idx); 36 | 37 | } // namespace caffe 38 | 39 | #endif // CAFFE_UTIL_HDF5_H_ 40 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/util/insert_splits.hpp: -------------------------------------------------------------------------------- 1 | #ifndef _CAFFE_UTIL_INSERT_SPLITS_HPP_ 2 | #define _CAFFE_UTIL_INSERT_SPLITS_HPP_ 3 | 4 | #include 5 | 6 | #include "caffe/proto/caffe.pb.h" 7 | 8 | namespace caffe { 9 | 10 | // Copy NetParameters with SplitLayers added to replace any shared bottom 11 | // blobs with unique bottom blobs provided by the SplitLayer. 12 | void InsertSplits(const NetParameter& param, NetParameter* param_split); 13 | 14 | void ConfigureSplitLayer(const string& layer_name, const string& blob_name, 15 | const int blob_idx, const int split_count, const float loss_weight, 16 | LayerParameter* split_layer_param); 17 | 18 | string SplitLayerName(const string& layer_name, const string& blob_name, 19 | const int blob_idx); 20 | 21 | string SplitBlobName(const string& layer_name, const string& blob_name, 22 | const int blob_idx, const int split_idx); 23 | 24 | } // namespace caffe 25 | 26 | #endif // CAFFE_UTIL_INSERT_SPLITS_HPP_ 27 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/util/matio_io.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_UTIL_MATIO_IO_H_ 2 | #define CAFFE_UTIL_MATIO_IO_H_ 3 | 4 | #include 5 | #include 6 | 7 | #include "google/protobuf/message.h" 8 | #include "caffe/blob.hpp" 9 | #include "caffe/common.hpp" 10 | #include "caffe/proto/caffe.pb.h" 11 | 12 | namespace caffe { 13 | 14 | using ::google::protobuf::Message; 15 | 16 | template 17 | void ReadBlobFromMat(const char *fname, Blob* blob); 18 | 19 | template 20 | void WriteBlobToMat(const char *fname, bool write_diff, 21 | Blob* blob); 22 | 23 | } // namespace caffe 24 | 25 | #endif // CAFFE_UTIL_IO_H_ 26 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/util/rng.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CAFFE_RNG_CPP_HPP_ 2 | #define CAFFE_RNG_CPP_HPP_ 3 | 4 | #include 5 | #include 6 | 7 | #include "boost/random/mersenne_twister.hpp" 8 | #include "boost/random/uniform_int.hpp" 9 | 10 | #include "caffe/common.hpp" 11 | 12 | namespace caffe { 13 | 14 | typedef boost::mt19937 rng_t; 15 | 16 | inline rng_t* caffe_rng() { 17 | return static_cast(Caffe::rng_stream().generator()); 18 | } 19 | 20 | // Fisher–Yates algorithm 21 | template 22 | inline void shuffle(RandomAccessIterator begin, RandomAccessIterator end, 23 | RandomGenerator* gen) { 24 | typedef typename std::iterator_traits::difference_type 25 | difference_type; 26 | typedef typename boost::uniform_int dist_type; 27 | 28 | difference_type length = std::distance(begin, end); 29 | if (length <= 0) return; 30 | 31 | for (difference_type i = length - 1; i > 0; --i) { 32 | dist_type dist(0, i); 33 | std::iter_swap(begin + i, begin + dist(*gen)); 34 | } 35 | } 36 | 37 | template 38 | inline void shuffle(RandomAccessIterator begin, RandomAccessIterator end) { 39 | shuffle(begin, end, caffe_rng()); 40 | } 41 | } // namespace caffe 42 | 43 | #endif // CAFFE_RNG_HPP_ 44 | -------------------------------------------------------------------------------- /deeplab-caffe/include/caffe/util/signal_handler.h: -------------------------------------------------------------------------------- 1 | #ifndef INCLUDE_CAFFE_UTIL_SIGNAL_HANDLER_H_ 2 | #define INCLUDE_CAFFE_UTIL_SIGNAL_HANDLER_H_ 3 | 4 | #include "caffe/proto/caffe.pb.h" 5 | #include "caffe/solver.hpp" 6 | 7 | namespace caffe { 8 | 9 | class SignalHandler { 10 | public: 11 | // Contructor. Specify what action to take when a signal is received. 12 | SignalHandler(SolverAction::Enum SIGINT_action, 13 | SolverAction::Enum SIGHUP_action); 14 | ~SignalHandler(); 15 | ActionCallback GetActionFunction(); 16 | private: 17 | SolverAction::Enum CheckForSignals() const; 18 | SolverAction::Enum SIGINT_action_; 19 | SolverAction::Enum SIGHUP_action_; 20 | }; 21 | 22 | } // namespace caffe 23 | 24 | #endif // INCLUDE_CAFFE_UTIL_SIGNAL_HANDLER_H_ 25 | -------------------------------------------------------------------------------- /deeplab-caffe/matlab/+caffe/+test/test_io.m: -------------------------------------------------------------------------------- 1 | classdef test_io < matlab.unittest.TestCase 2 | methods (Test) 3 | function test_read_write_mean(self) 4 | % randomly generate mean data 5 | width = 200; 6 | height = 300; 7 | channels = 3; 8 | mean_data_write = 255 * rand(width, height, channels, 'single'); 9 | % write mean data to binary proto 10 | mean_proto_file = tempname(); 11 | caffe.io.write_mean(mean_data_write, mean_proto_file); 12 | % read mean data from saved binary proto and test whether they are equal 13 | mean_data_read = caffe.io.read_mean(mean_proto_file); 14 | self.verifyEqual(mean_data_write, mean_data_read) 15 | delete(mean_proto_file); 16 | end 17 | end 18 | end 19 | -------------------------------------------------------------------------------- /deeplab-caffe/matlab/+caffe/+test/test_solver.m: -------------------------------------------------------------------------------- 1 | classdef test_solver < matlab.unittest.TestCase 2 | 3 | properties 4 | num_output 5 | solver 6 | end 7 | 8 | methods 9 | function self = test_solver() 10 | self.num_output = 13; 11 | model_file = caffe.test.test_net.simple_net_file(self.num_output); 12 | solver_file = tempname(); 13 | 14 | fid = fopen(solver_file, 'w'); 15 | fprintf(fid, [ ... 16 | 'net: "' model_file '"\n' ... 17 | 'test_iter: 10 test_interval: 10 base_lr: 0.01 momentum: 0.9\n' ... 18 | 'weight_decay: 0.0005 lr_policy: "inv" gamma: 0.0001 power: 0.75\n' ... 19 | 'display: 100 max_iter: 100 snapshot_after_train: false\n' ]); 20 | fclose(fid); 21 | 22 | self.solver = caffe.Solver(solver_file); 23 | % also make sure get_solver runs 24 | caffe.get_solver(solver_file); 25 | caffe.set_mode_cpu(); 26 | % fill in valid labels 27 | self.solver.net.blobs('label').set_data(randi( ... 28 | self.num_output - 1, self.solver.net.blobs('label').shape)); 29 | self.solver.test_nets(1).blobs('label').set_data(randi( ... 30 | self.num_output - 1, self.solver.test_nets(1).blobs('label').shape)); 31 | 32 | delete(solver_file); 33 | delete(model_file); 34 | end 35 | end 36 | methods (Test) 37 | function test_solve(self) 38 | self.verifyEqual(self.solver.iter(), 0) 39 | self.solver.step(30); 40 | self.verifyEqual(self.solver.iter(), 30) 41 | self.solver.solve() 42 | self.verifyEqual(self.solver.iter(), 100) 43 | end 44 | end 45 | end 46 | -------------------------------------------------------------------------------- /deeplab-caffe/matlab/+caffe/Layer.m: -------------------------------------------------------------------------------- 1 | classdef Layer < handle 2 | % Wrapper class of caffe::Layer in matlab 3 | 4 | properties (Access = private) 5 | hLayer_self 6 | attributes 7 | % attributes fields: 8 | % hBlob_blobs 9 | end 10 | properties (SetAccess = private) 11 | params 12 | end 13 | 14 | methods 15 | function self = Layer(hLayer_layer) 16 | CHECK(is_valid_handle(hLayer_layer), 'invalid Layer handle'); 17 | 18 | % setup self handle and attributes 19 | self.hLayer_self = hLayer_layer; 20 | self.attributes = caffe_('layer_get_attr', self.hLayer_self); 21 | 22 | % setup weights 23 | self.params = caffe.Blob.empty(); 24 | for n = 1:length(self.attributes.hBlob_blobs) 25 | self.params(n) = caffe.Blob(self.attributes.hBlob_blobs(n)); 26 | end 27 | end 28 | function layer_type = type(self) 29 | layer_type = caffe_('layer_get_type', self.hLayer_self); 30 | end 31 | end 32 | end 33 | -------------------------------------------------------------------------------- /deeplab-caffe/matlab/+caffe/get_net.m: -------------------------------------------------------------------------------- 1 | function net = get_net(varargin) 2 | % net = get_net(model_file, phase_name) or 3 | % net = get_net(model_file, weights_file, phase_name) 4 | % Construct a net from model_file, and load weights from weights_file 5 | % phase_name can only be 'train' or 'test' 6 | 7 | CHECK(nargin == 2 || nargin == 3, ['usage: ' ... 8 | 'net = get_net(model_file, phase_name) or ' ... 9 | 'net = get_net(model_file, weights_file, phase_name)']); 10 | if nargin == 3 11 | model_file = varargin{1}; 12 | weights_file = varargin{2}; 13 | phase_name = varargin{3}; 14 | elseif nargin == 2 15 | model_file = varargin{1}; 16 | phase_name = varargin{2}; 17 | end 18 | 19 | CHECK(ischar(model_file), 'model_file must be a string'); 20 | CHECK(ischar(phase_name), 'phase_name must be a string'); 21 | CHECK_FILE_EXIST(model_file); 22 | CHECK(strcmp(phase_name, 'train') || strcmp(phase_name, 'test'), ... 23 | sprintf('phase_name can only be %strain%s or %stest%s', ... 24 | char(39), char(39), char(39), char(39))); 25 | 26 | % construct caffe net from model_file 27 | hNet = caffe_('get_net', model_file, phase_name); 28 | net = caffe.Net(hNet); 29 | 30 | % load weights from weights_file 31 | if nargin == 3 32 | CHECK(ischar(weights_file), 'weights_file must be a string'); 33 | CHECK_FILE_EXIST(weights_file); 34 | net.copy_from(weights_file); 35 | end 36 | 37 | end 38 | -------------------------------------------------------------------------------- /deeplab-caffe/matlab/+caffe/get_solver.m: -------------------------------------------------------------------------------- 1 | function solver = get_solver(solver_file) 2 | % solver = get_solver(solver_file) 3 | % Construct a Solver object from solver_file 4 | 5 | CHECK(ischar(solver_file), 'solver_file must be a string'); 6 | CHECK_FILE_EXIST(solver_file); 7 | pSolver = caffe_('get_solver', solver_file); 8 | solver = caffe.Solver(pSolver); 9 | 10 | end 11 | -------------------------------------------------------------------------------- /deeplab-caffe/matlab/+caffe/imagenet/ilsvrc_2012_mean.mat: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/matlab/+caffe/imagenet/ilsvrc_2012_mean.mat -------------------------------------------------------------------------------- /deeplab-caffe/matlab/+caffe/private/CHECK.m: -------------------------------------------------------------------------------- 1 | function CHECK(expr, error_msg) 2 | 3 | if ~expr 4 | error(error_msg); 5 | end 6 | 7 | end 8 | -------------------------------------------------------------------------------- /deeplab-caffe/matlab/+caffe/private/CHECK_FILE_EXIST.m: -------------------------------------------------------------------------------- 1 | function CHECK_FILE_EXIST(filename) 2 | 3 | if exist(filename, 'file') == 0 4 | error('%s does not exist', filename); 5 | end 6 | 7 | end 8 | -------------------------------------------------------------------------------- /deeplab-caffe/matlab/+caffe/private/is_valid_handle.m: -------------------------------------------------------------------------------- 1 | function valid = is_valid_handle(hObj) 2 | % valid = is_valid_handle(hObj) or is_valid_handle('get_new_init_key') 3 | % Check if a handle is valid (has the right data type and init_key matches) 4 | % Use is_valid_handle('get_new_init_key') to get new init_key from C++; 5 | 6 | % a handle is a struct array with the following fields 7 | % (uint64) ptr : the pointer to the C++ object 8 | % (double) init_key : caffe initialization key 9 | 10 | persistent init_key; 11 | if isempty(init_key) 12 | init_key = caffe_('get_init_key'); 13 | end 14 | 15 | % is_valid_handle('get_new_init_key') to get new init_key from C++; 16 | if ischar(hObj) && strcmp(hObj, 'get_new_init_key') 17 | init_key = caffe_('get_init_key'); 18 | return 19 | else 20 | % check whether data types are correct and init_key matches 21 | valid = isstruct(hObj) ... 22 | && isscalar(hObj.ptr) && isa(hObj.ptr, 'uint64') ... 23 | && isscalar(hObj.init_key) && isa(hObj.init_key, 'double') ... 24 | && hObj.init_key == init_key; 25 | end 26 | 27 | end 28 | -------------------------------------------------------------------------------- /deeplab-caffe/matlab/+caffe/reset_all.m: -------------------------------------------------------------------------------- 1 | function reset_all() 2 | % reset_all() 3 | % clear all solvers and stand-alone nets and reset Caffe to initial status 4 | 5 | caffe_('reset'); 6 | is_valid_handle('get_new_init_key'); 7 | 8 | end 9 | -------------------------------------------------------------------------------- /deeplab-caffe/matlab/+caffe/run_tests.m: -------------------------------------------------------------------------------- 1 | function results = run_tests() 2 | % results = run_tests() 3 | % run all tests in this caffe matlab wrapper package 4 | 5 | % use CPU for testing 6 | caffe.set_mode_cpu(); 7 | 8 | % reset caffe before testing 9 | caffe.reset_all(); 10 | 11 | % put all test cases here 12 | results = [... 13 | run(caffe.test.test_net) ... 14 | run(caffe.test.test_solver) ... 15 | run(caffe.test.test_io) ]; 16 | 17 | % reset caffe after testing 18 | caffe.reset_all(); 19 | 20 | end 21 | -------------------------------------------------------------------------------- /deeplab-caffe/matlab/+caffe/set_device.m: -------------------------------------------------------------------------------- 1 | function set_device(device_id) 2 | % set_device(device_id) 3 | % set Caffe's GPU device ID 4 | 5 | CHECK(isscalar(device_id) && device_id >= 0, ... 6 | 'device_id must be non-negative integer'); 7 | device_id = double(device_id); 8 | 9 | caffe_('set_device', device_id); 10 | 11 | end 12 | -------------------------------------------------------------------------------- /deeplab-caffe/matlab/+caffe/set_mode_cpu.m: -------------------------------------------------------------------------------- 1 | function set_mode_cpu() 2 | % set_mode_cpu() 3 | % set Caffe to CPU mode 4 | 5 | caffe_('set_mode_cpu'); 6 | 7 | end 8 | -------------------------------------------------------------------------------- /deeplab-caffe/matlab/+caffe/set_mode_gpu.m: -------------------------------------------------------------------------------- 1 | function set_mode_gpu() 2 | % set_mode_gpu() 3 | % set Caffe to GPU mode 4 | 5 | caffe_('set_mode_gpu'); 6 | 7 | end 8 | -------------------------------------------------------------------------------- /deeplab-caffe/matlab/+caffe/version.m: -------------------------------------------------------------------------------- 1 | function version_str = version() 2 | % version() 3 | % show Caffe's version. 4 | 5 | version_str = caffe_('version'); 6 | 7 | end 8 | -------------------------------------------------------------------------------- /deeplab-caffe/matlab/hdf5creation/.gitignore: -------------------------------------------------------------------------------- 1 | *.h5 2 | list.txt 3 | -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/0_0_147_blob_0.mat: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/matlab/my_script/0_0_147_blob_0.mat -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/0_0_272.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/matlab/my_script/0_0_272.png -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/1.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/matlab/my_script/1.png -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/2007_000762.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/matlab/my_script/2007_000762.png -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/75.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/matlab/my_script/75.png -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/9990.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/matlab/my_script/9990.png -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/AppendPrefixPostfix.m: -------------------------------------------------------------------------------- 1 | function out = AppendPrefixPostfix(in, prefix, postfix) 2 | out = cell(size(in)); 3 | 4 | for i = 1 : numel(in) 5 | out{i} = [prefix, in{i}, postfix]; 6 | end 7 | end -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/EvalSaliencyResults.m: -------------------------------------------------------------------------------- 1 | SetupEnv; 2 | addpath('./funcs'); 3 | %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 4 | dataset_root_folder = sprintf('/home/phoenix/deeplab/rmt/data/%s',dataset); 5 | exper_root_folder = sprintf('/home/phoenix/deeplab/exper/%s',dataset); 6 | result_folder = sprintf('%s/results/%s/%s', exper_root_folder, model_name, testset); 7 | 8 | % Get Opts 9 | salopts = GetSaliencyOpts(dataset_root_folder, exper_root_folder, result_folder,trainset, testset, dataset) 10 | 11 | if strcmp(testset, 'val') 12 | [ conf, rawcounts] = MySaliencyEval(salopts, id); 13 | else 14 | fprintf(1, 'This is test set. No evaluation. Just saved as png\n'); 15 | end 16 | 17 | 18 | 19 | 20 | -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/GetImglistForCaffe.m: -------------------------------------------------------------------------------- 1 | %function GetImglistForCaffe() 2 | 3 | data_folder = '../VOC2012/ImageSets/Segmentation'; 4 | save_folder = '~/workspace/caffe-dev/examples/segnet'; 5 | 6 | img_prefix = '/JPEGImages/'; 7 | img_postfix = '.jpg'; 8 | seg_prefix = '/SegmentationClassAug/'; 9 | seg_postfix = '.png'; 10 | 11 | fn = {'VOC2012_test.txt', ... 12 | 'VOC2012_train.txt', ... 13 | 'VOC2012_val.txt', ... 14 | 'VOC2012_train_aug.txt', ... 15 | 'VOC2012_trainval_aug.txt'}; 16 | 17 | for i = 1 : numel(fn) 18 | list = GetList(fullfile(data_folder, fn{i})); 19 | imglist = AppendPrefixPostfix(list, img_prefix, img_postfix); 20 | seglist = AppendPrefixPostfix(list, seg_prefix, seg_postfix); 21 | 22 | assert(numel(imglist) == numel(seglist)); 23 | 24 | fid = fopen(fullfile(save_folder, fn{i}), 'w'); 25 | 26 | for j = 1 : numel(imglist) 27 | fprintf(fid, '%s %s\n', imglist{j}, seglist{j}); 28 | end 29 | fclose(fid); 30 | end 31 | 32 | 33 | 34 | -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/GetList.m: -------------------------------------------------------------------------------- 1 | function list = GetList(fn) 2 | fid2 = fopen(fn, 'r'); 3 | list = textscan(fid2, '%s'); 4 | list = list{1}; 5 | fclose(fid2); 6 | end -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/GetSaliencyOpts.m: -------------------------------------------------------------------------------- 1 | function salopts = GetSaliencyOpts(sal_data_dir, sal_exper_dir, sal_res_dir, trainset, testset, dataset) 2 | %clear msraopts 3 | 4 | if nargin < 5 5 | dataset = 'msra'; 6 | end 7 | 8 | % dataset 9 | salopts.dataset=dataset; 10 | 11 | 12 | % initialize the training set 13 | salopts.trainset = trainset; 14 | 15 | % initialize the test set 16 | salopts.testset = testset; 17 | 18 | % initialize main paths 19 | salopts.imgsetpath=[sal_exper_dir '/list/%s_id.txt']; 20 | salopts.respath = [sal_res_dir '/%s.png']; 21 | if strcmp(dataset, 'msra') || strcmp(dataset, 'msra10k') 22 | salopts.imgpath=[sal_data_dir '/msra/image/%s.jpg']; 23 | salopts.annopath=[sal_data_dir '/msra/gt/%s.png']; 24 | elseif strcmp(dataset, 'pascal-s') 25 | salopts.imgpath=[sal_data_dir '/imgs/%s.jpg']; 26 | salopts.annopath=[sal_data_dir '/gt/%s.png']; 27 | elseif strcmp(dataset, 'ECSSD') 28 | salopts.imgpath=[sal_data_dir '/image/%s.jpg']; 29 | salopts.annopath=[sal_data_dir '/gt/%s.png']; 30 | else 31 | error('Unknown Dataset!'); 32 | end 33 | % classes 34 | salopts.classes={... 35 | 'saliency' 36 | }; 37 | salopts.nclasses=length(salopts.classes); 38 | 39 | % overlap threshold 40 | salopts.minoverlap=0.5; 41 | 42 | 43 | -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/LoadBinFile.m: -------------------------------------------------------------------------------- 1 | function out = LoadBinFile(fn, type) 2 | % load binary file 3 | 4 | fid = fopen(fn, 'rb'); 5 | 6 | row = fread(fid, 1, 'int32'); 7 | col = fread(fid, 1, 'int32'); 8 | channel = fread(fid, 1, 'int32'); 9 | num = fread(fid, 1, 'int32'); 10 | 11 | num_ele = row*col*channel*num; 12 | 13 | if strcmp(type, 'int32') 14 | out = fread(fid, num_ele, 'int32'); 15 | elseif strcmp(type, 'single') 16 | out = fread(fid, num_ele, 'single'); 17 | elseif strcmp(type, 'uint8') 18 | out = fread(fid, num_ele, 'uint8'); 19 | else 20 | error('wrong type') 21 | end 22 | 23 | out = reshape(out, [row, col, channel, num]); 24 | 25 | fclose(fid); -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/SaveBinFile.m: -------------------------------------------------------------------------------- 1 | function SaveBinFile(data, fn, type) 2 | % save as binary file 3 | 4 | fid = fopen(fn, 'wb'); 5 | 6 | row = size(data, 1); 7 | col = size(data, 2); 8 | channel = size(data, 3); 9 | 10 | fwrite(fid, row, 'int32'); 11 | fwrite(fid, col, 'int32'); 12 | fwrite(fid, channel, 'int32'); 13 | 14 | if strcmp(type, 'double') 15 | fwrite(fid, data(:), 'double'); 16 | elseif strcmp(type, 'single') || strcmp(type, 'float') 17 | fwrite(fid, data(:), 'single'); 18 | elseif strcmp(type, 'uint8') 19 | fwrite(fid, data(:), 'uint8'); 20 | elseif strcmp(type, 'int32') 21 | fwrite(fid, data(:), 'int32'); 22 | else 23 | error('wrong type') 24 | end 25 | 26 | fclose(fid); -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/ShowGT.m: -------------------------------------------------------------------------------- 1 | tmp = load('pascal_seg_colormap.mat'); 2 | colormap = tmp.colormap; 3 | 4 | fn = '2007_000032.png'; 5 | gt1 = imread(fullfile('../SegmentationClass', fn)); 6 | gt2 = imread(fullfile('../SegmentationClassAug', fn)); 7 | figure(1), subplot(121), imshow(gt1, colormap), subplot(122), imshow(gt2, colormap) -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/SortSegmentationByBbox.m: -------------------------------------------------------------------------------- 1 | function [val ind] = SortSegmentationByBbox(segmentation, type) 2 | % sort instances within the segmentation map by area size 3 | % 4 | 5 | if nargin < 2 6 | type = 'descend'; 7 | end 8 | 9 | if ~strcmp(type, 'ascend') && ~strcmp(type, 'descend') 10 | error('wrong type\n') 11 | end 12 | 13 | labels = unique(segmentation(:)); 14 | 15 | areas = zeros(1, numel(labels)); 16 | 17 | for i = 1 : numel(labels) 18 | [row col] = find(segmentation == labels(i)); 19 | 20 | areas(i) = (max(row) - min(row)) * (max(col) - min(col)); 21 | end 22 | 23 | [val ind] = sort(areas, type); 24 | 25 | %ind = ind - 1; %instance id starts from 0 26 | -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/TransformBerkeleyVOC2011Annot.m: -------------------------------------------------------------------------------- 1 | clear all; close all; 2 | 3 | dataset = 'VOC2012'; 4 | 5 | %copy berkeley annotations 6 | orig_folder = '../Berkeley_annot/dataset/cls'; 7 | save_folder = ['../', dataset, '/SegmentationClassAug_Visualization']; 8 | 9 | if ~exist(save_folder, 'dir') 10 | mkdir(save_folder) 11 | end 12 | 13 | tmp = load('pascal_seg_colormap.mat'); 14 | colormap = tmp.colormap; 15 | 16 | annots = dir(fullfile(orig_folder, '*.mat')); 17 | 18 | for i = 1 : numel(annots) 19 | fprintf(1, 'processing %d (%d) ...\n', i, numel(annots)); 20 | 21 | gt = load(fullfile(orig_folder, annots(i).name)); 22 | 23 | imwrite(gt.GTcls.Segmentation, colormap, fullfile(save_folder, [annots(i).name(1:end-4), '.png'])); 24 | end 25 | 26 | % copy pascal annotations 27 | orig_folder = ['../', dataset, '/SegmentationClass']; 28 | annots = dir(fullfile(orig_folder, '*.png')); 29 | 30 | for i = 1 : numel(annots) 31 | fprintf(1, 'processing %d (%d) ...\n', i, numel(annots)); 32 | 33 | gt = imread(fullfile(orig_folder, annots(i).name)); 34 | 35 | imwrite(gt, colormap, fullfile(save_folder, annots(i).name)); 36 | end -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/cross_avgIOU_voc12_features_fc8DownSample118.txtcross_avgIOU_al_d_: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/matlab/my_script/cross_avgIOU_voc12_features_fc8DownSample118.txtcross_avgIOU_al_d_ -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/funcs/CalFmeasure.m: -------------------------------------------------------------------------------- 1 | function [ ratio ] = CalFmeasure(precision, recall, beta_square) 2 | ratio = (1+beta_square) * precision .* recall; 3 | denominator = beta_square*precision + recall; 4 | 5 | ratio = ratio ./ denominator; 6 | end -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/funcs/CalMAE.m: -------------------------------------------------------------------------------- 1 | function mae = CalMAE(smap, gtImg) 2 | % Code Author: Wangjiang Zhu 3 | % Email: wangjiang88119@gmail.com 4 | % Date: 3/24/2014 5 | if size(smap, 1) ~= size(gtImg, 1) || size(smap, 2) ~= size(gtImg, 2) 6 | error('Saliency map and gt Image have different sizes!\n'); 7 | end 8 | 9 | if ~islogical(gtImg) 10 | gtImg = gtImg(:,:,1) > 128; 11 | end 12 | 13 | smap = im2double(smap(:,:,1)); 14 | fgPixels = smap(gtImg); 15 | fgErrSum = length(fgPixels) - sum(fgPixels); 16 | bgErrSum = sum(smap(~gtImg)); 17 | mae = (fgErrSum + bgErrSum) / numel(gtImg); -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/funcs/CalMeanMAE.m: -------------------------------------------------------------------------------- 1 | function mae = CalMeanMAE(SRC, srcSuffix, GT, gtSuffix) 2 | % Code Author: Wangjiang Zhu 3 | % Email: wangjiang88119@gmail.com 4 | % Date: 3/24/2014 5 | files = dir(fullfile(SRC, strcat('*', srcSuffix))); 6 | if isempty(files) 7 | error('No saliency maps are found: %s\n', fullfile(SRC, strcat('*', srcSuffix))); 8 | end 9 | 10 | MAE = zeros(length(files), 1); 11 | parfor k = 1:length(files) 12 | srcName = files(k).name; 13 | srcImg = imread(fullfile(SRC, srcName)); 14 | 15 | gtName = strrep(srcName, srcSuffix, gtSuffix); 16 | gtImg = imread(fullfile(GT, gtName)); 17 | 18 | MAE(k) = CalMAE(srcImg, gtImg); 19 | end 20 | 21 | mae = mean(MAE); 22 | fprintf('MAE for %s: %f\n', srcSuffix, mae); -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/funcs/CalMeanPR.m: -------------------------------------------------------------------------------- 1 | function [pr,rc] = CalMeanPR(SRC, srcSuffix, GT, gtSuffix) 2 | % Code Author: Wangjiang Zhu 3 | % Email: wangjiang88119@gmail.com 4 | % Date: 3/24/2014 5 | files = dir(fullfile(SRC, strcat('*', srcSuffix))); 6 | if isempty(files) 7 | error('No saliency maps are found: %s\n', fullfile(SRC, strcat('*', srcSuffix))); 8 | end 9 | 10 | pr = zeros(length(files), 1); 11 | rc = zeros(length(files), 1); 12 | parfor k = 1:length(files) 13 | 14 | srcName = files(k).name; 15 | srcImg = imread(fullfile(SRC, srcName)); 16 | 17 | gtName = strrep(srcName, srcSuffix, gtSuffix); 18 | gtImg = imread(fullfile(GT, gtName)); 19 | 20 | ta = sum(srcImg(:))/(size(srcImg,1)*size(srcImg,2)); 21 | EVAL = Evaluate(gtImg(:,:,1)>1,srcImg>ta); 22 | 23 | pr(k) = EVAL(4); 24 | rc(k) = EVAL(5); 25 | end 26 | 27 | pr = mean(pr) 28 | rc = mean(rc) 29 | -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/funcs/CalPR.m: -------------------------------------------------------------------------------- 1 | function [precision, recall] = CalPR(smapImg, gtImg, targetIsFg, targetIsHigh) 2 | % Code Author: Wangjiang Zhu 3 | % Email: wangjiang88119@gmail.com 4 | % Date: 3/24/2014 5 | smapImg = smapImg(:,:,1); 6 | if ~islogical(gtImg) 7 | gtImg = gtImg(:,:,1) > 128; 8 | end 9 | if any(size(smapImg) ~= size(gtImg)) 10 | error('saliency map and ground truth mask have different size'); 11 | end 12 | 13 | if ~targetIsFg 14 | gtImg = ~gtImg; 15 | end 16 | 17 | 18 | gtPxlNum = sum(gtImg(:)); 19 | if 0 == gtPxlNum 20 | %error('no foreground region is labeled'); 21 | gtPxlNum = 1; 22 | end 23 | 24 | targetHist = histc(double(smapImg(gtImg)), 0:255); 25 | nontargetHist = histc(double(smapImg(~gtImg)), 0:255); 26 | 27 | if targetIsHigh 28 | targetHist = flipud(targetHist); 29 | nontargetHist = flipud(nontargetHist); 30 | end 31 | targetHist = cumsum( targetHist ); 32 | nontargetHist = cumsum( nontargetHist ); 33 | 34 | precision = targetHist ./ (targetHist + nontargetHist); 35 | if any(isnan(precision)) 36 | warning('there exists NAN in precision, this is because of your saliency map do not have a full range specified by cutThreshes\n'); 37 | end 38 | recall = targetHist / gtPxlNum; -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/pascal_seg_colormap.mat: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/matlab/my_script/pascal_seg_colormap.mat -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/result.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/matlab/my_script/result.png -------------------------------------------------------------------------------- /deeplab-caffe/matlab/my_script/saliency_colormap.mat: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/matlab/my_script/saliency_colormap.mat -------------------------------------------------------------------------------- /deeplab-caffe/python/caffe/__init__.py: -------------------------------------------------------------------------------- 1 | from .pycaffe import Net, SGDSolver, NesterovSolver, AdaGradSolver, RMSPropSolver, AdaDeltaSolver, AdamSolver 2 | from ._caffe import set_mode_cpu, set_mode_gpu, set_device, Layer, get_solver, layer_type_list 3 | from ._caffe import __version__ 4 | from .proto.caffe_pb2 import TRAIN, TEST 5 | from .classifier import Classifier 6 | from .detector import Detector 7 | from . import io 8 | from .net_spec import layers, params, NetSpec, to_proto 9 | -------------------------------------------------------------------------------- /deeplab-caffe/python/caffe/imagenet/ilsvrc_2012_mean.npy: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/python/caffe/imagenet/ilsvrc_2012_mean.npy -------------------------------------------------------------------------------- /deeplab-caffe/python/caffe/test/test_io.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import unittest 3 | 4 | import caffe 5 | 6 | class TestBlobProtoToArray(unittest.TestCase): 7 | 8 | def test_old_format(self): 9 | data = np.zeros((10,10)) 10 | blob = caffe.proto.caffe_pb2.BlobProto() 11 | blob.data.extend(list(data.flatten())) 12 | shape = (1,1,10,10) 13 | blob.num, blob.channels, blob.height, blob.width = shape 14 | 15 | arr = caffe.io.blobproto_to_array(blob) 16 | self.assertEqual(arr.shape, shape) 17 | 18 | def test_new_format(self): 19 | data = np.zeros((10,10)) 20 | blob = caffe.proto.caffe_pb2.BlobProto() 21 | blob.data.extend(list(data.flatten())) 22 | blob.shape.dim.extend(list(data.shape)) 23 | 24 | arr = caffe.io.blobproto_to_array(blob) 25 | self.assertEqual(arr.shape, data.shape) 26 | 27 | def test_no_shape(self): 28 | data = np.zeros((10,10)) 29 | blob = caffe.proto.caffe_pb2.BlobProto() 30 | blob.data.extend(list(data.flatten())) 31 | 32 | with self.assertRaises(ValueError): 33 | caffe.io.blobproto_to_array(blob) 34 | 35 | def test_scalar(self): 36 | data = np.ones((1)) * 123 37 | blob = caffe.proto.caffe_pb2.BlobProto() 38 | blob.data.extend(list(data.flatten())) 39 | 40 | arr = caffe.io.blobproto_to_array(blob) 41 | self.assertEqual(arr, 123) 42 | -------------------------------------------------------------------------------- /deeplab-caffe/python/caffe/test/test_layer_type_list.py: -------------------------------------------------------------------------------- 1 | import unittest 2 | 3 | import caffe 4 | 5 | class TestLayerTypeList(unittest.TestCase): 6 | 7 | def test_standard_types(self): 8 | #removing 'Data' from list 9 | for type_name in ['Data', 'Convolution', 'InnerProduct']: 10 | self.assertIn(type_name, caffe.layer_type_list(), 11 | '%s not in layer_type_list()' % type_name) 12 | -------------------------------------------------------------------------------- /deeplab-caffe/python/draw_net.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env python 2 | """ 3 | Draw a graph of the net architecture. 4 | """ 5 | from argparse import ArgumentParser, ArgumentDefaultsHelpFormatter 6 | from google.protobuf import text_format 7 | 8 | import caffe 9 | import caffe.draw 10 | from caffe.proto import caffe_pb2 11 | 12 | 13 | def parse_args(): 14 | """Parse input arguments 15 | """ 16 | 17 | parser = ArgumentParser(description=__doc__, 18 | formatter_class=ArgumentDefaultsHelpFormatter) 19 | 20 | parser.add_argument('input_net_proto_file', 21 | help='Input network prototxt file') 22 | parser.add_argument('output_image_file', 23 | help='Output image file') 24 | parser.add_argument('--rankdir', 25 | help=('One of TB (top-bottom, i.e., vertical), ' 26 | 'RL (right-left, i.e., horizontal), or another ' 27 | 'valid dot option; see ' 28 | 'http://www.graphviz.org/doc/info/' 29 | 'attrs.html#k:rankdir'), 30 | default='LR') 31 | 32 | args = parser.parse_args() 33 | return args 34 | 35 | 36 | def main(): 37 | args = parse_args() 38 | net = caffe_pb2.NetParameter() 39 | text_format.Merge(open(args.input_net_proto_file).read(), net) 40 | print('Drawing net to %s' % args.output_image_file) 41 | caffe.draw.draw_net_to_file(net, args.output_image_file, args.rankdir) 42 | 43 | 44 | if __name__ == '__main__': 45 | main() 46 | -------------------------------------------------------------------------------- /deeplab-caffe/python/requirements.txt: -------------------------------------------------------------------------------- 1 | Cython>=0.19.2 2 | numpy>=1.7.1 3 | scipy>=0.13.2 4 | scikit-image>=0.9.3 5 | matplotlib>=1.3.1 6 | ipython>=3.0.0 7 | h5py>=2.2.0 8 | leveldb>=0.191 9 | networkx>=1.8.1 10 | nose>=1.3.0 11 | pandas>=0.12.0 12 | python-dateutil>=1.4,<2 13 | protobuf>=2.5.0 14 | python-gflags>=2.0 15 | pyyaml>=3.10 16 | Pillow>=2.3.0 17 | six>=1.1.0 -------------------------------------------------------------------------------- /deeplab-caffe/scripts/build_docs.sh: -------------------------------------------------------------------------------- 1 | #!/bin/bash 2 | # Build documentation for display in web browser. 3 | 4 | PORT=${1:-4000} 5 | 6 | echo "usage: build_docs.sh [port]" 7 | 8 | # Find the docs dir, no matter where the script is called 9 | ROOT_DIR="$( cd "$(dirname "$0")"/.. ; pwd -P )" 10 | cd $ROOT_DIR 11 | 12 | # Gather docs. 13 | scripts/gather_examples.sh 14 | 15 | # Generate developer docs. 16 | make docs 17 | 18 | # Display docs using web server. 19 | cd docs 20 | jekyll serve -w -s . -d _site --port=$PORT 21 | -------------------------------------------------------------------------------- /deeplab-caffe/scripts/copy_notebook.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env python 2 | """ 3 | Takes as arguments: 4 | 1. the path to a JSON file (such as an IPython notebook). 5 | 2. the path to output file 6 | 7 | If 'metadata' dict in the JSON file contains 'include_in_docs': true, 8 | then copies the file to output file, appending the 'metadata' property 9 | as YAML front-matter, adding the field 'category' with value 'notebook'. 10 | """ 11 | import os 12 | import sys 13 | import json 14 | 15 | filename = sys.argv[1] 16 | output_filename = sys.argv[2] 17 | content = json.load(open(filename)) 18 | 19 | if 'include_in_docs' in content['metadata'] and content['metadata']['include_in_docs']: 20 | yaml_frontmatter = ['---'] 21 | for key, val in content['metadata'].iteritems(): 22 | if key == 'example_name': 23 | key = 'title' 24 | if val == '': 25 | val = os.path.basename(filename) 26 | yaml_frontmatter.append('{}: {}'.format(key, val)) 27 | yaml_frontmatter += ['category: notebook'] 28 | yaml_frontmatter += ['original_path: ' + filename] 29 | 30 | with open(output_filename, 'w') as fo: 31 | fo.write('\n'.join(yaml_frontmatter + ['---']) + '\n') 32 | fo.write(open(filename).read()) 33 | -------------------------------------------------------------------------------- /deeplab-caffe/scripts/download_model_from_gist.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env sh 2 | 3 | GIST=$1 4 | DIRNAME=${2:-./models} 5 | 6 | if [ -z $GIST ]; then 7 | echo "usage: download_model_from_gist.sh " 8 | exit 9 | fi 10 | 11 | GIST_DIR=$(echo $GIST | tr '/' '-') 12 | MODEL_DIR="$DIRNAME/$GIST_DIR" 13 | 14 | if [ -d $MODEL_DIR ]; then 15 | echo "$MODEL_DIR already exists! Please make sure you're not overwriting anything important!" 16 | exit 17 | fi 18 | 19 | echo "Downloading Caffe model info to $MODEL_DIR ..." 20 | mkdir -p $MODEL_DIR 21 | wget https://gist.github.com/$GIST/download -O $MODEL_DIR/gist.zip 22 | unzip -j $MODEL_DIR/gist.zip -d $MODEL_DIR 23 | rm $MODEL_DIR/gist.zip 24 | echo "Done" 25 | -------------------------------------------------------------------------------- /deeplab-caffe/scripts/gather_examples.sh: -------------------------------------------------------------------------------- 1 | #!/bin/bash 2 | # Assemble documentation for the project into one directory via symbolic links. 3 | 4 | # Find the docs dir, no matter where the script is called 5 | ROOT_DIR="$( cd "$(dirname "$0")"/.. ; pwd -P )" 6 | cd $ROOT_DIR 7 | 8 | # Gather docs from examples/**/readme.md 9 | GATHERED_DIR=docs/gathered 10 | rm -r $GATHERED_DIR 11 | mkdir $GATHERED_DIR 12 | for README_FILENAME in $(find examples -iname "readme.md"); do 13 | # Only use file if it is to be included in docs. 14 | if grep -Fxq "include_in_docs: true" $README_FILENAME; then 15 | # Make link to readme.md in docs/gathered/. 16 | # Since everything is called readme.md, rename it by its dirname. 17 | README_DIRNAME=`dirname $README_FILENAME` 18 | DOCS_FILENAME=$GATHERED_DIR/$README_DIRNAME.md 19 | mkdir -p `dirname $DOCS_FILENAME` 20 | ln -s $ROOT_DIR/$README_FILENAME $DOCS_FILENAME 21 | fi 22 | done 23 | 24 | # Gather docs from examples/*.ipynb and add YAML front-matter. 25 | for NOTEBOOK_FILENAME in $(find examples -depth -iname "*.ipynb"); do 26 | DOCS_FILENAME=$GATHERED_DIR/$NOTEBOOK_FILENAME 27 | mkdir -p `dirname $DOCS_FILENAME` 28 | python scripts/copy_notebook.py $NOTEBOOK_FILENAME $DOCS_FILENAME 29 | done 30 | -------------------------------------------------------------------------------- /deeplab-caffe/scripts/travis/travis_build_and_test.sh: -------------------------------------------------------------------------------- 1 | #!/bin/bash 2 | # Script called by Travis to build and test Caffe. 3 | # Travis CI tests are CPU-only for lack of compatible hardware. 4 | 5 | set -e 6 | MAKE="make --jobs=$NUM_THREADS --keep-going" 7 | 8 | if $WITH_CMAKE; then 9 | mkdir build 10 | cd build 11 | CPU_ONLY=" -DCPU_ONLY=ON" 12 | if ! $WITH_CUDA; then 13 | CPU_ONLY=" -DCPU_ONLY=OFF" 14 | fi 15 | PYTHON_ARGS="" 16 | if [ "$PYTHON_VERSION" = "3" ]; then 17 | PYTHON_ARGS="$PYTHON_ARGS -Dpython_version=3 -DBOOST_LIBRARYDIR=$CONDA_DIR/lib/" 18 | fi 19 | if $WITH_IO; then 20 | IO_ARGS="-DUSE_OPENCV=ON -DUSE_LMDB=ON -DUSE_LEVELDB=ON" 21 | else 22 | IO_ARGS="-DUSE_OPENCV=OFF -DUSE_LMDB=OFF -DUSE_LEVELDB=OFF" 23 | fi 24 | cmake -DBUILD_python=ON -DCMAKE_BUILD_TYPE=Release $CPU_ONLY $PYTHON_ARGS -DCMAKE_INCLUDE_PATH="$CONDA_DIR/include/" -DCMAKE_LIBRARY_PATH="$CONDA_DIR/lib/" $IO_ARGS .. 25 | $MAKE 26 | $MAKE pytest 27 | if ! $WITH_CUDA; then 28 | $MAKE runtest 29 | $MAKE lint 30 | fi 31 | $MAKE clean 32 | cd - 33 | else 34 | if ! $WITH_CUDA; then 35 | export CPU_ONLY=1 36 | fi 37 | if $WITH_IO; then 38 | export USE_LMDB=1 39 | export USE_LEVELDB=1 40 | export USE_OPENCV=1 41 | fi 42 | $MAKE all test pycaffe warn lint || true 43 | if ! $WITH_CUDA; then 44 | $MAKE runtest 45 | fi 46 | $MAKE all 47 | $MAKE test 48 | $MAKE pycaffe 49 | $MAKE pytest 50 | $MAKE warn 51 | if ! $WITH_CUDA; then 52 | $MAKE lint 53 | fi 54 | fi 55 | -------------------------------------------------------------------------------- /deeplab-caffe/scripts/travis/travis_setup_makefile_config.sh: -------------------------------------------------------------------------------- 1 | #!/bin/bash 2 | 3 | set -e 4 | 5 | mv Makefile.config.example Makefile.config 6 | 7 | if $WITH_CUDA; then 8 | # Only generate compute_50. 9 | GENCODE="-gencode arch=compute_50,code=sm_50" 10 | GENCODE="$GENCODE -gencode arch=compute_50,code=compute_50" 11 | echo "CUDA_ARCH := $GENCODE" >> Makefile.config 12 | fi 13 | 14 | # Remove IO library settings from Makefile.config 15 | # to avoid conflicts with CI configuration 16 | sed -i -e '/USE_LMDB/d' Makefile.config 17 | sed -i -e '/USE_LEVELDB/d' Makefile.config 18 | sed -i -e '/USE_OPENCV/d' Makefile.config 19 | 20 | cat << 'EOF' >> Makefile.config 21 | # Travis' nvcc doesn't like newer boost versions 22 | NVCCFLAGS := -Xcudafe --diag_suppress=cc_clobber_ignored -Xcudafe --diag_suppress=useless_using_declaration -Xcudafe --diag_suppress=set_but_not_used 23 | ANACONDA_HOME := $(CONDA_DIR) 24 | PYTHON_INCLUDE := $(ANACONDA_HOME)/include \ 25 | $(ANACONDA_HOME)/include/python2.7 \ 26 | $(ANACONDA_HOME)/lib/python2.7/site-packages/numpy/core/include 27 | PYTHON_LIB := $(ANACONDA_HOME)/lib 28 | INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include 29 | LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib 30 | WITH_PYTHON_LAYER := 1 31 | EOF 32 | -------------------------------------------------------------------------------- /deeplab-caffe/scripts/upload_model_to_gist.sh: -------------------------------------------------------------------------------- 1 | #!/bin/bash 2 | 3 | # Check for valid directory 4 | DIRNAME=$1 5 | if [ ! -f $DIRNAME/readme.md ]; then 6 | echo "usage: upload_model_to_gist.sh " 7 | echo " /readme.md must exist" 8 | fi 9 | cd $DIRNAME 10 | FILES=`find . -maxdepth 1 -type f ! -name "*.caffemodel*" | xargs echo` 11 | 12 | # Check for gist tool. 13 | gist -v >/dev/null 2>&1 || { echo >&2 "I require 'gist' but it's not installed. Do 'gem install gist'."; exit 1; } 14 | 15 | NAME=`sed -n 's/^name:[[:space:]]*//p' readme.md` 16 | if [ -z "$NAME" ]; then 17 | echo " /readme.md must contain name field in the front-matter." 18 | fi 19 | 20 | GIST=`sed -n 's/^gist_id:[[:space:]]*//p' readme.md` 21 | if [ -z "$GIST" ]; then 22 | echo "Uploading new Gist" 23 | gist -p -d "$NAME" $FILES 24 | else 25 | echo "Updating existing Gist, id $GIST" 26 | gist -u $GIST -d "$NAME" $FILES 27 | fi 28 | 29 | RESULT=$? 30 | if [ $RESULT -eq 0 ]; then 31 | echo "You've uploaded your model!" 32 | echo "Don't forget to add the gist_id field to your /readme.md now!" 33 | echo "Run the command again after you do that, to make sure the Gist id propagates." 34 | echo "" 35 | echo "And do share your model over at https://github.com/BVLC/caffe/wiki/Model-Zoo" 36 | else 37 | echo "Something went wrong!" 38 | fi 39 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/CMakeLists.txt: -------------------------------------------------------------------------------- 1 | # generate protobuf sources 2 | file(GLOB proto_files proto/*.proto) 3 | caffe_protobuf_generate_cpp_py(${proto_gen_folder} proto_srcs proto_hdrs proto_python ${proto_files}) 4 | 5 | # include python files either to force generation 6 | add_library(proto STATIC ${proto_hdrs} ${proto_srcs} ${proto_python}) 7 | set(Caffe_LINKER_LIBS proto ${Caffe_LINKER_LIBS}) # note, crucial to prepend! 8 | caffe_default_properties(proto) 9 | 10 | # --[ Caffe library 11 | 12 | # creates 'test_srcs', 'srcs', 'test_cuda', 'cuda' lists 13 | caffe_pickup_caffe_sources(${PROJECT_SOURCE_DIR}) 14 | 15 | if(HAVE_CUDA) 16 | caffe_cuda_compile(cuda_objs ${cuda}) 17 | list(APPEND srcs ${cuda_objs} ${cuda}) 18 | endif() 19 | 20 | add_library(caffe ${srcs}) 21 | target_link_libraries(caffe proto ${Caffe_LINKER_LIBS}) 22 | caffe_default_properties(caffe) 23 | set_target_properties(caffe PROPERTIES 24 | VERSION ${CAFFE_TARGET_VERSION} 25 | SOVERSION ${CAFFE_TARGET_SOVERSION} 26 | ) 27 | 28 | # ---[ Tests 29 | add_subdirectory(test) 30 | 31 | # ---[ Install 32 | install(DIRECTORY ${Caffe_INCLUDE_DIR}/caffe DESTINATION include) 33 | install(FILES ${proto_hdrs} DESTINATION include/caffe/proto) 34 | install(TARGETS caffe proto EXPORT CaffeTargets DESTINATION lib) 35 | 36 | file(WRITE ${PROJECT_BINARY_DIR}/__init__.py) 37 | list(APPEND proto_python ${PROJECT_BINARY_DIR}/__init__.py) 38 | install(PROGRAMS ${proto_python} DESTINATION python/caffe/proto) 39 | 40 | 41 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layer.cpp: -------------------------------------------------------------------------------- 1 | #include 2 | #include "caffe/layer.hpp" 3 | 4 | namespace caffe { 5 | 6 | template 7 | void Layer::InitMutex() { 8 | forward_mutex_.reset(new boost::mutex()); 9 | } 10 | 11 | template 12 | void Layer::Lock() { 13 | if (IsShared()) { 14 | forward_mutex_->lock(); 15 | } 16 | } 17 | 18 | template 19 | void Layer::Unlock() { 20 | if (IsShared()) { 21 | forward_mutex_->unlock(); 22 | } 23 | } 24 | 25 | INSTANTIATE_CLASS(Layer); 26 | 27 | } // namespace caffe 28 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/absval_layer.cpp: -------------------------------------------------------------------------------- 1 | #include 2 | 3 | #include "caffe/layers/absval_layer.hpp" 4 | #include "caffe/util/math_functions.hpp" 5 | 6 | namespace caffe { 7 | 8 | template 9 | void AbsValLayer::LayerSetUp(const vector*>& bottom, 10 | const vector*>& top) { 11 | NeuronLayer::LayerSetUp(bottom, top); 12 | CHECK_NE(top[0], bottom[0]) << this->type() << " Layer does not " 13 | "allow in-place computation."; 14 | } 15 | 16 | template 17 | void AbsValLayer::Forward_cpu( 18 | const vector*>& bottom, const vector*>& top) { 19 | const int count = top[0]->count(); 20 | Dtype* top_data = top[0]->mutable_cpu_data(); 21 | caffe_abs(count, bottom[0]->cpu_data(), top_data); 22 | } 23 | 24 | template 25 | void AbsValLayer::Backward_cpu(const vector*>& top, 26 | const vector& propagate_down, const vector*>& bottom) { 27 | const int count = top[0]->count(); 28 | const Dtype* top_diff = top[0]->cpu_diff(); 29 | if (propagate_down[0]) { 30 | const Dtype* bottom_data = bottom[0]->cpu_data(); 31 | Dtype* bottom_diff = bottom[0]->mutable_cpu_diff(); 32 | caffe_cpu_sign(count, bottom_data, bottom_diff); 33 | caffe_mul(count, bottom_diff, top_diff, bottom_diff); 34 | } 35 | } 36 | 37 | #ifdef CPU_ONLY 38 | STUB_GPU(AbsValLayer); 39 | #endif 40 | 41 | INSTANTIATE_CLASS(AbsValLayer); 42 | REGISTER_LAYER_CLASS(AbsVal); 43 | 44 | } // namespace caffe 45 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/absval_layer.cu: -------------------------------------------------------------------------------- 1 | #include 2 | 3 | #include "caffe/layers/absval_layer.hpp" 4 | #include "caffe/util/math_functions.hpp" 5 | 6 | namespace caffe { 7 | 8 | template 9 | void AbsValLayer::Forward_gpu( 10 | const vector*>& bottom, const vector*>& top) { 11 | const int count = top[0]->count(); 12 | Dtype* top_data = top[0]->mutable_gpu_data(); 13 | caffe_gpu_abs(count, bottom[0]->gpu_data(), top_data); 14 | } 15 | 16 | template 17 | void AbsValLayer::Backward_gpu(const vector*>& top, 18 | const vector& propagate_down, const vector*>& bottom) { 19 | const int count = top[0]->count(); 20 | const Dtype* top_diff = top[0]->gpu_diff(); 21 | if (propagate_down[0]) { 22 | const Dtype* bottom_data = bottom[0]->gpu_data(); 23 | Dtype* bottom_diff = bottom[0]->mutable_gpu_diff(); 24 | caffe_gpu_sign(count, bottom_data, bottom_diff); 25 | caffe_gpu_mul(count, bottom_diff, top_diff, bottom_diff); 26 | } 27 | } 28 | 29 | INSTANTIATE_LAYER_GPU_FUNCS(AbsValLayer); 30 | 31 | 32 | } // namespace caffe 33 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/bnll_layer.cpp: -------------------------------------------------------------------------------- 1 | #include 2 | #include 3 | 4 | #include "caffe/layers/bnll_layer.hpp" 5 | 6 | namespace caffe { 7 | 8 | const float kBNLL_THRESHOLD = 50.; 9 | 10 | template 11 | void BNLLLayer::Forward_cpu(const vector*>& bottom, 12 | const vector*>& top) { 13 | const Dtype* bottom_data = bottom[0]->cpu_data(); 14 | Dtype* top_data = top[0]->mutable_cpu_data(); 15 | const int count = bottom[0]->count(); 16 | for (int i = 0; i < count; ++i) { 17 | top_data[i] = bottom_data[i] > 0 ? 18 | bottom_data[i] + log(1. + exp(-bottom_data[i])) : 19 | log(1. + exp(bottom_data[i])); 20 | } 21 | } 22 | 23 | template 24 | void BNLLLayer::Backward_cpu(const vector*>& top, 25 | const vector& propagate_down, 26 | const vector*>& bottom) { 27 | if (propagate_down[0]) { 28 | const Dtype* bottom_data = bottom[0]->cpu_data(); 29 | const Dtype* top_diff = top[0]->cpu_diff(); 30 | Dtype* bottom_diff = bottom[0]->mutable_cpu_diff(); 31 | const int count = bottom[0]->count(); 32 | Dtype expval; 33 | for (int i = 0; i < count; ++i) { 34 | expval = exp(std::min(bottom_data[i], Dtype(kBNLL_THRESHOLD))); 35 | bottom_diff[i] = top_diff[i] * expval / (expval + 1.); 36 | } 37 | } 38 | } 39 | 40 | #ifdef CPU_ONLY 41 | STUB_GPU(BNLLLayer); 42 | #endif 43 | 44 | INSTANTIATE_CLASS(BNLLLayer); 45 | REGISTER_LAYER_CLASS(BNLL); 46 | 47 | } // namespace caffe 48 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/cudnn_lrn_layer.cu: -------------------------------------------------------------------------------- 1 | #ifdef USE_CUDNN 2 | #include 3 | 4 | #include "caffe/layers/cudnn_lrn_layer.hpp" 5 | 6 | namespace caffe { 7 | 8 | template 9 | void CuDNNLRNLayer::Forward_gpu(const vector*>& bottom, 10 | const vector*>& top) { 11 | const Dtype* bottom_data = bottom[0]->gpu_data(); 12 | Dtype* top_data = top[0]->mutable_gpu_data(); 13 | 14 | CUDNN_CHECK(cudnnLRNCrossChannelForward( 15 | handle_, norm_desc_, CUDNN_LRN_CROSS_CHANNEL_DIM1, 16 | cudnn::dataType::one, 17 | bottom_desc_, bottom_data, 18 | cudnn::dataType::zero, 19 | top_desc_, top_data) ); 20 | } 21 | 22 | template 23 | void CuDNNLRNLayer::Backward_gpu(const vector*>& top, 24 | const vector& propagate_down, const vector*>& bottom) { 25 | const Dtype* top_diff = top[0]->gpu_diff(); 26 | const Dtype* top_data = top[0]->gpu_data(); 27 | const Dtype* bottom_data = bottom[0]->gpu_data(); 28 | Dtype* bottom_diff = bottom[0]->mutable_gpu_diff(); 29 | 30 | CUDNN_CHECK(cudnnLRNCrossChannelBackward( 31 | handle_, norm_desc_, CUDNN_LRN_CROSS_CHANNEL_DIM1, 32 | cudnn::dataType::one, 33 | top_desc_, top_data, 34 | top_desc_, top_diff, 35 | bottom_desc_, bottom_data, 36 | cudnn::dataType::zero, 37 | bottom_desc_, bottom_diff) ); 38 | } 39 | 40 | INSTANTIATE_LAYER_GPU_FUNCS(CuDNNLRNLayer); 41 | 42 | }; // namespace caffe 43 | 44 | #endif 45 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/cudnn_pooling_layer.cu: -------------------------------------------------------------------------------- 1 | #ifdef USE_CUDNN 2 | #include 3 | 4 | #include "caffe/layers/cudnn_pooling_layer.hpp" 5 | 6 | namespace caffe { 7 | 8 | template 9 | void CuDNNPoolingLayer::Forward_gpu(const vector*>& bottom, 10 | const vector*>& top) { 11 | const Dtype* bottom_data = bottom[0]->gpu_data(); 12 | Dtype* top_data = top[0]->mutable_gpu_data(); 13 | CUDNN_CHECK(cudnnPoolingForward(handle_, pooling_desc_, 14 | cudnn::dataType::one, 15 | bottom_desc_, bottom_data, 16 | cudnn::dataType::zero, 17 | top_desc_, top_data)); 18 | } 19 | 20 | template 21 | void CuDNNPoolingLayer::Backward_gpu(const vector*>& top, 22 | const vector& propagate_down, const vector*>& bottom) { 23 | if (!propagate_down[0]) { 24 | return; 25 | } 26 | const Dtype* top_diff = top[0]->gpu_diff(); 27 | const Dtype* top_data = top[0]->gpu_data(); 28 | const Dtype* bottom_data = bottom[0]->gpu_data(); 29 | Dtype* bottom_diff = bottom[0]->mutable_gpu_diff(); 30 | CUDNN_CHECK(cudnnPoolingBackward(handle_, pooling_desc_, 31 | cudnn::dataType::one, 32 | top_desc_, top_data, top_desc_, top_diff, 33 | bottom_desc_, bottom_data, 34 | cudnn::dataType::zero, 35 | bottom_desc_, bottom_diff)); 36 | } 37 | 38 | INSTANTIATE_LAYER_GPU_FUNCS(CuDNNPoolingLayer); 39 | 40 | } // namespace caffe 41 | #endif 42 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/cudnn_relu_layer.cpp: -------------------------------------------------------------------------------- 1 | #ifdef USE_CUDNN 2 | #include 3 | 4 | #include "caffe/layers/cudnn_relu_layer.hpp" 5 | 6 | namespace caffe { 7 | 8 | template 9 | void CuDNNReLULayer::LayerSetUp(const vector*>& bottom, 10 | const vector*>& top) { 11 | ReLULayer::LayerSetUp(bottom, top); 12 | // initialize cuDNN 13 | CUDNN_CHECK(cudnnCreate(&handle_)); 14 | cudnn::createTensor4dDesc(&bottom_desc_); 15 | cudnn::createTensor4dDesc(&top_desc_); 16 | handles_setup_ = true; 17 | } 18 | 19 | template 20 | void CuDNNReLULayer::Reshape(const vector*>& bottom, 21 | const vector*>& top) { 22 | ReLULayer::Reshape(bottom, top); 23 | const int N = bottom[0]->num(); 24 | const int K = bottom[0]->channels(); 25 | const int H = bottom[0]->height(); 26 | const int W = bottom[0]->width(); 27 | cudnn::setTensor4dDesc(&bottom_desc_, N, K, H, W); 28 | cudnn::setTensor4dDesc(&top_desc_, N, K, H, W); 29 | } 30 | 31 | template 32 | CuDNNReLULayer::~CuDNNReLULayer() { 33 | // Check that handles have been setup before destroying. 34 | if (!handles_setup_) { return; } 35 | 36 | cudnnDestroyTensorDescriptor(this->bottom_desc_); 37 | cudnnDestroyTensorDescriptor(this->top_desc_); 38 | cudnnDestroy(this->handle_); 39 | } 40 | 41 | INSTANTIATE_CLASS(CuDNNReLULayer); 42 | 43 | } // namespace caffe 44 | #endif 45 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/cudnn_sigmoid_layer.cpp: -------------------------------------------------------------------------------- 1 | #ifdef USE_CUDNN 2 | #include 3 | 4 | #include "caffe/layers/cudnn_sigmoid_layer.hpp" 5 | 6 | namespace caffe { 7 | 8 | template 9 | void CuDNNSigmoidLayer::LayerSetUp(const vector*>& bottom, 10 | const vector*>& top) { 11 | SigmoidLayer::LayerSetUp(bottom, top); 12 | // initialize cuDNN 13 | CUDNN_CHECK(cudnnCreate(&handle_)); 14 | cudnn::createTensor4dDesc(&bottom_desc_); 15 | cudnn::createTensor4dDesc(&top_desc_); 16 | handles_setup_ = true; 17 | } 18 | 19 | template 20 | void CuDNNSigmoidLayer::Reshape(const vector*>& bottom, 21 | const vector*>& top) { 22 | SigmoidLayer::Reshape(bottom, top); 23 | const int N = bottom[0]->num(); 24 | const int K = bottom[0]->channels(); 25 | const int H = bottom[0]->height(); 26 | const int W = bottom[0]->width(); 27 | cudnn::setTensor4dDesc(&bottom_desc_, N, K, H, W); 28 | cudnn::setTensor4dDesc(&top_desc_, N, K, H, W); 29 | } 30 | 31 | template 32 | CuDNNSigmoidLayer::~CuDNNSigmoidLayer() { 33 | // Check that handles have been setup before destroying. 34 | if (!handles_setup_) { return; } 35 | 36 | cudnnDestroyTensorDescriptor(this->bottom_desc_); 37 | cudnnDestroyTensorDescriptor(this->top_desc_); 38 | cudnnDestroy(this->handle_); 39 | } 40 | 41 | INSTANTIATE_CLASS(CuDNNSigmoidLayer); 42 | 43 | } // namespace caffe 44 | #endif 45 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/cudnn_sigmoid_layer.cu: -------------------------------------------------------------------------------- 1 | #ifdef USE_CUDNN 2 | #include 3 | 4 | #include "caffe/layers/cudnn_sigmoid_layer.hpp" 5 | 6 | namespace caffe { 7 | 8 | template 9 | void CuDNNSigmoidLayer::Forward_gpu(const vector*>& bottom, 10 | const vector*>& top) { 11 | const Dtype* bottom_data = bottom[0]->gpu_data(); 12 | Dtype* top_data = top[0]->mutable_gpu_data(); 13 | CUDNN_CHECK(cudnnActivationForward(this->handle_, 14 | CUDNN_ACTIVATION_SIGMOID, 15 | cudnn::dataType::one, 16 | this->bottom_desc_, bottom_data, 17 | cudnn::dataType::zero, 18 | this->top_desc_, top_data)); 19 | } 20 | 21 | template 22 | void CuDNNSigmoidLayer::Backward_gpu(const vector*>& top, 23 | const vector& propagate_down, 24 | const vector*>& bottom) { 25 | if (!propagate_down[0]) { 26 | return; 27 | } 28 | 29 | const Dtype* top_data = top[0]->gpu_data(); 30 | const Dtype* top_diff = top[0]->gpu_diff(); 31 | const Dtype* bottom_data = bottom[0]->gpu_data(); 32 | Dtype* bottom_diff = bottom[0]->mutable_gpu_diff(); 33 | CUDNN_CHECK(cudnnActivationBackward(this->handle_, 34 | CUDNN_ACTIVATION_SIGMOID, 35 | cudnn::dataType::one, 36 | this->top_desc_, top_data, this->top_desc_, top_diff, 37 | this->bottom_desc_, bottom_data, 38 | cudnn::dataType::zero, 39 | this->bottom_desc_, bottom_diff)); 40 | } 41 | 42 | INSTANTIATE_LAYER_GPU_FUNCS(CuDNNSigmoidLayer); 43 | 44 | } // namespace caffe 45 | #endif 46 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/cudnn_softmax_layer.cpp: -------------------------------------------------------------------------------- 1 | #ifdef USE_CUDNN 2 | #include 3 | 4 | #include "thrust/device_vector.h" 5 | 6 | #include "caffe/layers/cudnn_softmax_layer.hpp" 7 | 8 | namespace caffe { 9 | 10 | template 11 | void CuDNNSoftmaxLayer::LayerSetUp(const vector*>& bottom, 12 | const vector*>& top) { 13 | SoftmaxLayer::LayerSetUp(bottom, top); 14 | // Initialize CUDNN. 15 | CUDNN_CHECK(cudnnCreate(&handle_)); 16 | cudnn::createTensor4dDesc(&bottom_desc_); 17 | cudnn::createTensor4dDesc(&top_desc_); 18 | handles_setup_ = true; 19 | } 20 | 21 | template 22 | void CuDNNSoftmaxLayer::Reshape(const vector*>& bottom, 23 | const vector*>& top) { 24 | SoftmaxLayer::Reshape(bottom, top); 25 | int N = this->outer_num_; 26 | int K = bottom[0]->shape(this->softmax_axis_); 27 | int H = this->inner_num_; 28 | int W = 1; 29 | cudnn::setTensor4dDesc(&bottom_desc_, N, K, H, W); 30 | cudnn::setTensor4dDesc(&top_desc_, N, K, H, W); 31 | } 32 | 33 | template 34 | CuDNNSoftmaxLayer::~CuDNNSoftmaxLayer() { 35 | // Check that handles have been setup before destroying. 36 | if (!handles_setup_) { return; } 37 | 38 | cudnnDestroyTensorDescriptor(bottom_desc_); 39 | cudnnDestroyTensorDescriptor(top_desc_); 40 | cudnnDestroy(handle_); 41 | } 42 | 43 | INSTANTIATE_CLASS(CuDNNSoftmaxLayer); 44 | 45 | } // namespace caffe 46 | #endif 47 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/cudnn_softmax_layer.cu: -------------------------------------------------------------------------------- 1 | #ifdef USE_CUDNN 2 | #include 3 | 4 | #include "thrust/device_vector.h" 5 | 6 | #include "caffe/layers/cudnn_softmax_layer.hpp" 7 | 8 | namespace caffe { 9 | 10 | template 11 | void CuDNNSoftmaxLayer::Forward_gpu(const vector*>& bottom, 12 | const vector*>& top) { 13 | const Dtype* bottom_data = bottom[0]->gpu_data(); 14 | Dtype* top_data = top[0]->mutable_gpu_data(); 15 | CUDNN_CHECK(cudnnSoftmaxForward(handle_, CUDNN_SOFTMAX_ACCURATE, 16 | CUDNN_SOFTMAX_MODE_CHANNEL, 17 | cudnn::dataType::one, 18 | bottom_desc_, bottom_data, 19 | cudnn::dataType::zero, 20 | top_desc_, top_data)); 21 | } 22 | 23 | template 24 | void CuDNNSoftmaxLayer::Backward_gpu(const vector*>& top, 25 | const vector& propagate_down, const vector*>& bottom) { 26 | if (propagate_down[0]) { 27 | const Dtype* top_data = top[0]->gpu_data(); 28 | const Dtype* top_diff = top[0]->gpu_diff(); 29 | const Dtype* bottom_data = bottom[0]->gpu_data(); 30 | Dtype* bottom_diff = bottom[0]->mutable_gpu_diff(); 31 | 32 | CUDNN_CHECK(cudnnSoftmaxBackward(handle_, CUDNN_SOFTMAX_ACCURATE, 33 | CUDNN_SOFTMAX_MODE_CHANNEL, 34 | cudnn::dataType::one, 35 | top_desc_, top_data, top_desc_, top_diff, 36 | cudnn::dataType::zero, 37 | bottom_desc_, bottom_diff)); 38 | } 39 | } 40 | 41 | INSTANTIATE_LAYER_GPU_FUNCS(CuDNNSoftmaxLayer); 42 | 43 | } // namespace caffe 44 | #endif 45 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/cudnn_tanh_layer.cpp: -------------------------------------------------------------------------------- 1 | #ifdef USE_CUDNN 2 | #include 3 | 4 | #include "caffe/layers/cudnn_tanh_layer.hpp" 5 | 6 | namespace caffe { 7 | 8 | template 9 | void CuDNNTanHLayer::LayerSetUp(const vector*>& bottom, 10 | const vector*>& top) { 11 | TanHLayer::LayerSetUp(bottom, top); 12 | // initialize cuDNN 13 | CUDNN_CHECK(cudnnCreate(&handle_)); 14 | cudnn::createTensor4dDesc(&bottom_desc_); 15 | cudnn::createTensor4dDesc(&top_desc_); 16 | handles_setup_ = true; 17 | } 18 | 19 | template 20 | void CuDNNTanHLayer::Reshape(const vector*>& bottom, 21 | const vector*>& top) { 22 | TanHLayer::Reshape(bottom, top); 23 | const int N = bottom[0]->num(); 24 | const int K = bottom[0]->channels(); 25 | const int H = bottom[0]->height(); 26 | const int W = bottom[0]->width(); 27 | cudnn::setTensor4dDesc(&bottom_desc_, N, K, H, W); 28 | cudnn::setTensor4dDesc(&top_desc_, N, K, H, W); 29 | } 30 | 31 | template 32 | CuDNNTanHLayer::~CuDNNTanHLayer() { 33 | // Check that handles have been setup before destroying. 34 | if (!handles_setup_) { return; } 35 | 36 | cudnnDestroyTensorDescriptor(this->bottom_desc_); 37 | cudnnDestroyTensorDescriptor(this->top_desc_); 38 | cudnnDestroy(this->handle_); 39 | } 40 | 41 | INSTANTIATE_CLASS(CuDNNTanHLayer); 42 | 43 | } // namespace caffe 44 | #endif 45 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/cudnn_tanh_layer.cu: -------------------------------------------------------------------------------- 1 | #ifdef USE_CUDNN 2 | #include 3 | 4 | #include "caffe/layers/cudnn_tanh_layer.hpp" 5 | 6 | namespace caffe { 7 | 8 | template 9 | void CuDNNTanHLayer::Forward_gpu(const vector*>& bottom, 10 | const vector*>& top) { 11 | const Dtype* bottom_data = bottom[0]->gpu_data(); 12 | Dtype* top_data = top[0]->mutable_gpu_data(); 13 | CUDNN_CHECK(cudnnActivationForward(this->handle_, 14 | CUDNN_ACTIVATION_TANH, 15 | cudnn::dataType::one, 16 | this->bottom_desc_, bottom_data, 17 | cudnn::dataType::zero, 18 | this->top_desc_, top_data)); 19 | } 20 | 21 | template 22 | void CuDNNTanHLayer::Backward_gpu(const vector*>& top, 23 | const vector& propagate_down, 24 | const vector*>& bottom) { 25 | if (!propagate_down[0]) { 26 | return; 27 | } 28 | 29 | const Dtype* top_data = top[0]->gpu_data(); 30 | const Dtype* top_diff = top[0]->gpu_diff(); 31 | const Dtype* bottom_data = bottom[0]->gpu_data(); 32 | Dtype* bottom_diff = bottom[0]->mutable_gpu_diff(); 33 | 34 | CUDNN_CHECK(cudnnActivationBackward(this->handle_, 35 | CUDNN_ACTIVATION_TANH, 36 | cudnn::dataType::one, 37 | this->top_desc_, top_data, this->top_desc_, top_diff, 38 | this->bottom_desc_, bottom_data, 39 | cudnn::dataType::zero, 40 | this->bottom_desc_, bottom_diff)); 41 | } 42 | 43 | INSTANTIATE_LAYER_GPU_FUNCS(CuDNNTanHLayer); 44 | 45 | } // namespace caffe 46 | #endif 47 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/elu_layer.cpp: -------------------------------------------------------------------------------- 1 | #include 2 | #include 3 | 4 | #include "caffe/layers/elu_layer.hpp" 5 | 6 | namespace caffe { 7 | 8 | template 9 | void ELULayer::Forward_cpu(const vector*>& bottom, 10 | const vector*>& top) { 11 | const Dtype* bottom_data = bottom[0]->cpu_data(); 12 | Dtype* top_data = top[0]->mutable_cpu_data(); 13 | const int count = bottom[0]->count(); 14 | Dtype alpha = this->layer_param_.elu_param().alpha(); 15 | for (int i = 0; i < count; ++i) { 16 | top_data[i] = std::max(bottom_data[i], Dtype(0)) 17 | + alpha * (exp(std::min(bottom_data[i], Dtype(0))) - Dtype(1)); 18 | } 19 | } 20 | 21 | template 22 | void ELULayer::Backward_cpu(const vector*>& top, 23 | const vector& propagate_down, 24 | const vector*>& bottom) { 25 | if (propagate_down[0]) { 26 | const Dtype* bottom_data = bottom[0]->cpu_data(); 27 | const Dtype* top_data = top[0]->cpu_data(); 28 | const Dtype* top_diff = top[0]->cpu_diff(); 29 | Dtype* bottom_diff = bottom[0]->mutable_cpu_diff(); 30 | const int count = bottom[0]->count(); 31 | Dtype alpha = this->layer_param_.elu_param().alpha(); 32 | for (int i = 0; i < count; ++i) { 33 | bottom_diff[i] = top_diff[i] * ((bottom_data[i] > 0) 34 | + (alpha + top_data[i]) * (bottom_data[i] <= 0)); 35 | } 36 | } 37 | } 38 | 39 | 40 | #ifdef CPU_ONLY 41 | STUB_GPU(ELULayer); 42 | #endif 43 | 44 | INSTANTIATE_CLASS(ELULayer); 45 | REGISTER_LAYER_CLASS(ELU); 46 | 47 | } // namespace caffe 48 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/euclidean_loss_layer.cu: -------------------------------------------------------------------------------- 1 | #include 2 | 3 | #include "caffe/layers/euclidean_loss_layer.hpp" 4 | #include "caffe/util/math_functions.hpp" 5 | 6 | namespace caffe { 7 | 8 | template 9 | void EuclideanLossLayer::Forward_gpu(const vector*>& bottom, 10 | const vector*>& top) { 11 | int count = bottom[0]->count(); 12 | caffe_gpu_sub( 13 | count, 14 | bottom[0]->gpu_data(), 15 | bottom[1]->gpu_data(), 16 | diff_.mutable_gpu_data()); 17 | Dtype dot; 18 | caffe_gpu_dot(count, diff_.gpu_data(), diff_.gpu_data(), &dot); 19 | Dtype loss = dot / bottom[0]->num() / Dtype(2); 20 | top[0]->mutable_cpu_data()[0] = loss; 21 | } 22 | 23 | template 24 | void EuclideanLossLayer::Backward_gpu(const vector*>& top, 25 | const vector& propagate_down, const vector*>& bottom) { 26 | for (int i = 0; i < 2; ++i) { 27 | if (propagate_down[i]) { 28 | const Dtype sign = (i == 0) ? 1 : -1; 29 | const Dtype alpha = sign * top[0]->cpu_diff()[0] / bottom[i]->num(); 30 | caffe_gpu_axpby( 31 | bottom[i]->count(), // count 32 | alpha, // alpha 33 | diff_.gpu_data(), // a 34 | Dtype(0), // beta 35 | bottom[i]->mutable_gpu_diff()); // b 36 | } 37 | } 38 | } 39 | 40 | INSTANTIATE_LAYER_GPU_FUNCS(EuclideanLossLayer); 41 | 42 | } // namespace caffe 43 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/exp_layer.cu: -------------------------------------------------------------------------------- 1 | #include 2 | 3 | #include "caffe/layers/exp_layer.hpp" 4 | #include "caffe/util/math_functions.hpp" 5 | 6 | namespace caffe { 7 | 8 | template 9 | void ExpLayer::Forward_gpu(const vector*>& bottom, 10 | const vector*>& top) { 11 | const int count = bottom[0]->count(); 12 | const Dtype* bottom_data = bottom[0]->gpu_data(); 13 | Dtype* top_data = top[0]->mutable_gpu_data(); 14 | if (inner_scale_ == Dtype(1)) { 15 | caffe_gpu_exp(count, bottom_data, top_data); 16 | } else { 17 | caffe_gpu_scale(count, inner_scale_, bottom_data, top_data); 18 | caffe_gpu_exp(count, top_data, top_data); 19 | } 20 | if (outer_scale_ != Dtype(1)) { 21 | caffe_gpu_scal(count, outer_scale_, top_data); 22 | } 23 | } 24 | 25 | template 26 | void ExpLayer::Backward_gpu(const vector*>& top, 27 | const vector& propagate_down, const vector*>& bottom) { 28 | if (!propagate_down[0]) { return; } 29 | const int count = bottom[0]->count(); 30 | const Dtype* top_data = top[0]->gpu_data(); 31 | const Dtype* top_diff = top[0]->gpu_diff(); 32 | Dtype* bottom_diff = bottom[0]->mutable_gpu_diff(); 33 | caffe_gpu_mul(count, top_data, top_diff, bottom_diff); 34 | if (inner_scale_ != Dtype(1)) { 35 | caffe_gpu_scal(count, inner_scale_, bottom_diff); 36 | } 37 | } 38 | 39 | INSTANTIATE_LAYER_GPU_FUNCS(ExpLayer); 40 | 41 | 42 | } // namespace caffe 43 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/get_data_dim_layer.cpp: -------------------------------------------------------------------------------- 1 | #include 2 | 3 | #include "caffe/layers/get_data_dim_layer.hpp" 4 | 5 | namespace caffe { 6 | 7 | template 8 | void GetDataDimLayer::LayerSetUp( 9 | const vector*>& bottom, const vector*>& top) { 10 | CHECK_GE(bottom.size(), 1) 11 | << "Only accept one bottom"; 12 | } 13 | 14 | template 15 | void GetDataDimLayer::Reshape(const vector*>& bottom, 16 | const vector*>& top) { 17 | const int n = bottom[0]->num(); 18 | top[0]->Reshape(n, 1, 1, 2); 19 | } 20 | 21 | 22 | template 23 | void GetDataDimLayer::Forward_cpu(const vector*>& bottom, 24 | const vector*>& top) { 25 | // const Dtype* bottom_data = bottom[0]->cpu_data(); 26 | Dtype* top_data = top[0]->mutable_cpu_data(); 27 | 28 | const int h = bottom[0]->height(); 29 | const int w = bottom[0]->width(); 30 | const int n = bottom[0]->num(); 31 | 32 | for (int i = 0; i < n; ++i) { 33 | int top_data_offset = top[0]->offset(i); 34 | top_data[top_data_offset] = h; 35 | top_data[top_data_offset+1] = w; 36 | } 37 | } 38 | 39 | template 40 | void GetDataDimLayer::Backward_cpu(const vector*>& top, 41 | const vector& propagate_down, 42 | const vector*>& bottom) { 43 | } 44 | 45 | #ifdef CPU_ONLY 46 | STUB_GPU(GetDataDimLayer); 47 | #endif 48 | 49 | INSTANTIATE_CLASS(GetDataDimLayer); 50 | REGISTER_LAYER_CLASS(GetDataDim); 51 | 52 | } // namespace caffe 53 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/hdf5_output_layer.cu: -------------------------------------------------------------------------------- 1 | #include 2 | 3 | #include "hdf5.h" 4 | #include "hdf5_hl.h" 5 | 6 | #include "caffe/layers/hdf5_output_layer.hpp" 7 | 8 | namespace caffe { 9 | 10 | template 11 | void HDF5OutputLayer::Forward_gpu(const vector*>& bottom, 12 | const vector*>& top) { 13 | CHECK_GE(bottom.size(), 2); 14 | CHECK_EQ(bottom[0]->num(), bottom[1]->num()); 15 | data_blob_.Reshape(bottom[0]->num(), bottom[0]->channels(), 16 | bottom[0]->height(), bottom[0]->width()); 17 | label_blob_.Reshape(bottom[1]->num(), bottom[1]->channels(), 18 | bottom[1]->height(), bottom[1]->width()); 19 | const int data_datum_dim = bottom[0]->count() / bottom[0]->num(); 20 | const int label_datum_dim = bottom[1]->count() / bottom[1]->num(); 21 | 22 | for (int i = 0; i < bottom[0]->num(); ++i) { 23 | caffe_copy(data_datum_dim, &bottom[0]->gpu_data()[i * data_datum_dim], 24 | &data_blob_.mutable_cpu_data()[i * data_datum_dim]); 25 | caffe_copy(label_datum_dim, &bottom[1]->gpu_data()[i * label_datum_dim], 26 | &label_blob_.mutable_cpu_data()[i * label_datum_dim]); 27 | } 28 | SaveBlobs(); 29 | } 30 | 31 | template 32 | void HDF5OutputLayer::Backward_gpu(const vector*>& top, 33 | const vector& propagate_down, const vector*>& bottom) { 34 | return; 35 | } 36 | 37 | INSTANTIATE_LAYER_GPU_FUNCS(HDF5OutputLayer); 38 | 39 | } // namespace caffe 40 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/loss_layer.cpp: -------------------------------------------------------------------------------- 1 | #include 2 | 3 | #include "caffe/layers/loss_layer.hpp" 4 | 5 | namespace caffe { 6 | 7 | template 8 | void LossLayer::LayerSetUp( 9 | const vector*>& bottom, const vector*>& top) { 10 | // LossLayers have a non-zero (1) loss by default. 11 | if (this->layer_param_.loss_weight_size() == 0) { 12 | this->layer_param_.add_loss_weight(Dtype(1)); 13 | } 14 | } 15 | 16 | template 17 | void LossLayer::Reshape( 18 | const vector*>& bottom, const vector*>& top) { 19 | CHECK_EQ(bottom[0]->num(), bottom[1]->num()) 20 | << "The data and label should have the same number."; 21 | vector loss_shape(0); // Loss layers output a scalar; 0 axes. 22 | top[0]->Reshape(loss_shape); 23 | } 24 | 25 | INSTANTIATE_CLASS(LossLayer); 26 | 27 | } // namespace caffe 28 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/neuron_layer.cpp: -------------------------------------------------------------------------------- 1 | #include 2 | 3 | #include "caffe/layers/neuron_layer.hpp" 4 | 5 | namespace caffe { 6 | 7 | template 8 | void NeuronLayer::Reshape(const vector*>& bottom, 9 | const vector*>& top) { 10 | top[0]->ReshapeLike(*bottom[0]); 11 | } 12 | 13 | INSTANTIATE_CLASS(NeuronLayer); 14 | 15 | } // namespace caffe 16 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/relu_layer.cpp: -------------------------------------------------------------------------------- 1 | #include 2 | #include 3 | 4 | #include "caffe/layers/relu_layer.hpp" 5 | 6 | namespace caffe { 7 | 8 | template 9 | void ReLULayer::Forward_cpu(const vector*>& bottom, 10 | const vector*>& top) { 11 | const Dtype* bottom_data = bottom[0]->cpu_data(); 12 | Dtype* top_data = top[0]->mutable_cpu_data(); 13 | const int count = bottom[0]->count(); 14 | Dtype negative_slope = this->layer_param_.relu_param().negative_slope(); 15 | for (int i = 0; i < count; ++i) { 16 | top_data[i] = std::max(bottom_data[i], Dtype(0)) 17 | + negative_slope * std::min(bottom_data[i], Dtype(0)); 18 | } 19 | } 20 | 21 | template 22 | void ReLULayer::Backward_cpu(const vector*>& top, 23 | const vector& propagate_down, 24 | const vector*>& bottom) { 25 | if (propagate_down[0]) { 26 | const Dtype* bottom_data = bottom[0]->cpu_data(); 27 | const Dtype* top_diff = top[0]->cpu_diff(); 28 | Dtype* bottom_diff = bottom[0]->mutable_cpu_diff(); 29 | const int count = bottom[0]->count(); 30 | Dtype negative_slope = this->layer_param_.relu_param().negative_slope(); 31 | for (int i = 0; i < count; ++i) { 32 | bottom_diff[i] = top_diff[i] * ((bottom_data[i] > 0) 33 | + negative_slope * (bottom_data[i] <= 0)); 34 | } 35 | } 36 | } 37 | 38 | 39 | #ifdef CPU_ONLY 40 | STUB_GPU(ReLULayer); 41 | #endif 42 | 43 | INSTANTIATE_CLASS(ReLULayer); 44 | 45 | } // namespace caffe 46 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/sigmoid_layer.cpp: -------------------------------------------------------------------------------- 1 | #include 2 | #include 3 | 4 | #include "caffe/layers/sigmoid_layer.hpp" 5 | 6 | namespace caffe { 7 | 8 | template 9 | inline Dtype sigmoid(Dtype x) { 10 | return 1. / (1. + exp(-x)); 11 | } 12 | 13 | template 14 | void SigmoidLayer::Forward_cpu(const vector*>& bottom, 15 | const vector*>& top) { 16 | const Dtype* bottom_data = bottom[0]->cpu_data(); 17 | Dtype* top_data = top[0]->mutable_cpu_data(); 18 | const int count = bottom[0]->count(); 19 | for (int i = 0; i < count; ++i) { 20 | top_data[i] = sigmoid(bottom_data[i]); 21 | } 22 | } 23 | 24 | template 25 | void SigmoidLayer::Backward_cpu(const vector*>& top, 26 | const vector& propagate_down, 27 | const vector*>& bottom) { 28 | if (propagate_down[0]) { 29 | const Dtype* top_data = top[0]->cpu_data(); 30 | const Dtype* top_diff = top[0]->cpu_diff(); 31 | Dtype* bottom_diff = bottom[0]->mutable_cpu_diff(); 32 | const int count = bottom[0]->count(); 33 | for (int i = 0; i < count; ++i) { 34 | const Dtype sigmoid_x = top_data[i]; 35 | bottom_diff[i] = top_diff[i] * sigmoid_x * (1. - sigmoid_x); 36 | } 37 | } 38 | } 39 | 40 | #ifdef CPU_ONLY 41 | STUB_GPU(SigmoidLayer); 42 | #endif 43 | 44 | INSTANTIATE_CLASS(SigmoidLayer); 45 | 46 | 47 | } // namespace caffe 48 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/silence_layer.cpp: -------------------------------------------------------------------------------- 1 | #include 2 | 3 | #include "caffe/layers/silence_layer.hpp" 4 | #include "caffe/util/math_functions.hpp" 5 | 6 | namespace caffe { 7 | 8 | template 9 | void SilenceLayer::Backward_cpu(const vector*>& top, 10 | const vector& propagate_down, const vector*>& bottom) { 11 | for (int i = 0; i < bottom.size(); ++i) { 12 | if (propagate_down[i]) { 13 | caffe_set(bottom[i]->count(), Dtype(0), 14 | bottom[i]->mutable_cpu_diff()); 15 | } 16 | } 17 | } 18 | 19 | #ifdef CPU_ONLY 20 | STUB_GPU(SilenceLayer); 21 | #endif 22 | 23 | INSTANTIATE_CLASS(SilenceLayer); 24 | REGISTER_LAYER_CLASS(Silence); 25 | 26 | } // namespace caffe 27 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/silence_layer.cu: -------------------------------------------------------------------------------- 1 | #include 2 | 3 | #include "caffe/layers/silence_layer.hpp" 4 | #include "caffe/util/math_functions.hpp" 5 | 6 | namespace caffe { 7 | 8 | template 9 | void SilenceLayer::Forward_gpu(const vector*>& bottom, 10 | const vector*>& top) { 11 | // Do nothing. 12 | } 13 | 14 | template 15 | void SilenceLayer::Backward_gpu(const vector*>& top, 16 | const vector& propagate_down, const vector*>& bottom) { 17 | for (int i = 0; i < bottom.size(); ++i) { 18 | if (propagate_down[i]) { 19 | caffe_gpu_set(bottom[i]->count(), Dtype(0), 20 | bottom[i]->mutable_gpu_diff()); 21 | } 22 | } 23 | } 24 | 25 | INSTANTIATE_LAYER_GPU_FUNCS(SilenceLayer); 26 | 27 | } // namespace caffe 28 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/split_layer.cu: -------------------------------------------------------------------------------- 1 | #include 2 | 3 | #include "caffe/layers/split_layer.hpp" 4 | #include "caffe/util/math_functions.hpp" 5 | 6 | namespace caffe { 7 | 8 | template 9 | void SplitLayer::Forward_gpu(const vector*>& bottom, 10 | const vector*>& top) { 11 | for (int i = 0; i < top.size(); ++i) { 12 | top[i]->ShareData(*bottom[0]); 13 | } 14 | } 15 | 16 | template 17 | void SplitLayer::Backward_gpu(const vector*>& top, 18 | const vector& propagate_down, const vector*>& bottom) { 19 | if (!propagate_down[0]) { return; } 20 | if (top.size() == 1) { 21 | caffe_copy(count_, top[0]->gpu_diff(), bottom[0]->mutable_gpu_diff()); 22 | return; 23 | } 24 | caffe_gpu_add(count_, top[0]->gpu_diff(), top[1]->gpu_diff(), 25 | bottom[0]->mutable_gpu_diff()); 26 | // Add remaining top blob diffs. 27 | for (int i = 2; i < top.size(); ++i) { 28 | const Dtype* top_diff = top[i]->gpu_diff(); 29 | Dtype* bottom_diff = bottom[0]->mutable_gpu_diff(); 30 | caffe_gpu_axpy(count_, Dtype(1.), top_diff, bottom_diff); 31 | } 32 | } 33 | 34 | 35 | INSTANTIATE_LAYER_GPU_FUNCS(SplitLayer); 36 | 37 | } // namespace caffe 38 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/tanh_layer.cpp: -------------------------------------------------------------------------------- 1 | // TanH neuron activation function layer. 2 | // Adapted from ReLU layer code written by Yangqing Jia 3 | 4 | #include 5 | 6 | #include "caffe/layers/tanh_layer.hpp" 7 | 8 | namespace caffe { 9 | 10 | template 11 | void TanHLayer::Forward_cpu(const vector*>& bottom, 12 | const vector*>& top) { 13 | const Dtype* bottom_data = bottom[0]->cpu_data(); 14 | Dtype* top_data = top[0]->mutable_cpu_data(); 15 | const int count = bottom[0]->count(); 16 | for (int i = 0; i < count; ++i) { 17 | top_data[i] = tanh(bottom_data[i]); 18 | } 19 | } 20 | 21 | template 22 | void TanHLayer::Backward_cpu(const vector*>& top, 23 | const vector& propagate_down, 24 | const vector*>& bottom) { 25 | if (propagate_down[0]) { 26 | const Dtype* top_data = top[0]->cpu_data(); 27 | const Dtype* top_diff = top[0]->cpu_diff(); 28 | Dtype* bottom_diff = bottom[0]->mutable_cpu_diff(); 29 | const int count = bottom[0]->count(); 30 | Dtype tanhx; 31 | for (int i = 0; i < count; ++i) { 32 | tanhx = top_data[i]; 33 | bottom_diff[i] = top_diff[i] * (1 - tanhx * tanhx); 34 | } 35 | } 36 | } 37 | 38 | #ifdef CPU_ONLY 39 | STUB_GPU(TanHLayer); 40 | #endif 41 | 42 | INSTANTIATE_CLASS(TanHLayer); 43 | 44 | } // namespace caffe 45 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/threshold_layer.cpp: -------------------------------------------------------------------------------- 1 | #include 2 | 3 | #include "caffe/layers/threshold_layer.hpp" 4 | 5 | namespace caffe { 6 | 7 | template 8 | void ThresholdLayer::LayerSetUp(const vector*>& bottom, 9 | const vector*>& top) { 10 | NeuronLayer::LayerSetUp(bottom, top); 11 | threshold_ = this->layer_param_.threshold_param().threshold(); 12 | } 13 | 14 | template 15 | void ThresholdLayer::Forward_cpu(const vector*>& bottom, 16 | const vector*>& top) { 17 | const Dtype* bottom_data = bottom[0]->cpu_data(); 18 | Dtype* top_data = top[0]->mutable_cpu_data(); 19 | const int count = bottom[0]->count(); 20 | for (int i = 0; i < count; ++i) { 21 | top_data[i] = (bottom_data[i] > threshold_) ? Dtype(1) : Dtype(0); 22 | } 23 | } 24 | 25 | #ifdef CPU_ONLY 26 | STUB_GPU_FORWARD(ThresholdLayer, Forward); 27 | #endif 28 | 29 | INSTANTIATE_CLASS(ThresholdLayer); 30 | REGISTER_LAYER_CLASS(Threshold); 31 | 32 | } // namespace caffe 33 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/layers/threshold_layer.cu: -------------------------------------------------------------------------------- 1 | #include 2 | 3 | #include "caffe/layers/threshold_layer.hpp" 4 | 5 | namespace caffe { 6 | 7 | template 8 | __global__ void ThresholdForward(const int n, const Dtype threshold, 9 | const Dtype* in, Dtype* out) { 10 | CUDA_KERNEL_LOOP(index, n) { 11 | out[index] = in[index] > threshold ? 1 : 0; 12 | } 13 | } 14 | 15 | template 16 | void ThresholdLayer::Forward_gpu(const vector*>& bottom, 17 | const vector*>& top) { 18 | const Dtype* bottom_data = bottom[0]->gpu_data(); 19 | Dtype* top_data = top[0]->mutable_gpu_data(); 20 | const int count = bottom[0]->count(); 21 | // NOLINT_NEXT_LINE(whitespace/operators) 22 | ThresholdForward<<>>( 23 | count, threshold_, bottom_data, top_data); 24 | CUDA_POST_KERNEL_CHECK; 25 | } 26 | 27 | 28 | INSTANTIATE_LAYER_GPU_FORWARD(ThresholdLayer); 29 | 30 | 31 | } // namespace caffe 32 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/solvers/adadelta_solver.cu: -------------------------------------------------------------------------------- 1 | #include "caffe/util/math_functions.hpp" 2 | 3 | 4 | namespace caffe { 5 | 6 | template 7 | __global__ void AdaDeltaUpdate(int N, Dtype* g, Dtype* h, Dtype* h2, 8 | Dtype momentum, Dtype delta, Dtype local_rate) { 9 | CUDA_KERNEL_LOOP(i, N) { 10 | float gi = g[i]; 11 | float hi = h[i] = momentum * h[i] + (1-momentum) * gi * gi; 12 | gi = gi * sqrt((h2[i] + delta) / (hi + delta)); 13 | h2[i] = momentum * h2[i] + (1-momentum) * gi * gi; 14 | g[i] = local_rate * gi; 15 | } 16 | } 17 | template 18 | void adadelta_update_gpu(int N, Dtype* g, Dtype* h, Dtype* h2, Dtype momentum, 19 | Dtype delta, Dtype local_rate) { 20 | AdaDeltaUpdate // NOLINT_NEXT_LINE(whitespace/operators) 21 | <<>>( 22 | N, g, h, h2, momentum, delta, local_rate); 23 | CUDA_POST_KERNEL_CHECK; 24 | } 25 | template void adadelta_update_gpu(int , float*, float*, float*, 26 | float, float, float); 27 | template void adadelta_update_gpu(int, double*, double*, double*, 28 | double, double, double); 29 | 30 | } // namespace caffe 31 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/solvers/adagrad_solver.cu: -------------------------------------------------------------------------------- 1 | #include "caffe/util/math_functions.hpp" 2 | 3 | 4 | namespace caffe { 5 | 6 | template 7 | __global__ void AdaGradUpdate(int N, Dtype* g, Dtype* h, Dtype delta, 8 | Dtype local_rate) { 9 | CUDA_KERNEL_LOOP(i, N) { 10 | float gi = g[i]; 11 | float hi = h[i] = h[i] + gi*gi; 12 | g[i] = local_rate * gi / (sqrt(hi) + delta); 13 | } 14 | } 15 | template 16 | void adagrad_update_gpu(int N, Dtype* g, Dtype* h, Dtype delta, 17 | Dtype local_rate) { 18 | AdaGradUpdate // NOLINT_NEXT_LINE(whitespace/operators) 19 | <<>>( 20 | N, g, h, delta, local_rate); 21 | CUDA_POST_KERNEL_CHECK; 22 | } 23 | template void adagrad_update_gpu(int, float*, float*, float, float); 24 | template void adagrad_update_gpu(int, double*, double*, double, double); 25 | 26 | } // namespace caffe 27 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/solvers/adam_solver.cu: -------------------------------------------------------------------------------- 1 | #include "caffe/util/math_functions.hpp" 2 | 3 | 4 | namespace caffe { 5 | 6 | template 7 | __global__ void AdamUpdate(int N, Dtype* g, Dtype* m, Dtype* v, 8 | Dtype beta1, Dtype beta2, Dtype eps_hat, Dtype corrected_local_rate) { 9 | CUDA_KERNEL_LOOP(i, N) { 10 | float gi = g[i]; 11 | float mi = m[i] = m[i]*beta1 + gi*(1-beta1); 12 | float vi = v[i] = v[i]*beta2 + gi*gi*(1-beta2); 13 | g[i] = corrected_local_rate * mi / (sqrt(vi) + eps_hat); 14 | } 15 | } 16 | template 17 | void adam_update_gpu(int N, Dtype* g, Dtype* m, Dtype* v, Dtype beta1, 18 | Dtype beta2, Dtype eps_hat, Dtype corrected_local_rate) { 19 | AdamUpdate // NOLINT_NEXT_LINE(whitespace/operators) 20 | <<>>( 21 | N, g, m, v, beta1, beta2, eps_hat, corrected_local_rate); 22 | CUDA_POST_KERNEL_CHECK; 23 | } 24 | template void adam_update_gpu(int, float*, float*, float*, 25 | float, float, float, float); 26 | template void adam_update_gpu(int, double*, double*, double*, 27 | double, double, double, double); 28 | 29 | } // namespace caffe 30 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/solvers/nesterov_solver.cu: -------------------------------------------------------------------------------- 1 | #include "caffe/util/math_functions.hpp" 2 | 3 | 4 | namespace caffe { 5 | 6 | template 7 | __global__ void NesterovUpdate(int N, Dtype* g, Dtype* h, 8 | Dtype momentum, Dtype local_rate) { 9 | CUDA_KERNEL_LOOP(i, N) { 10 | float hi = h[i]; 11 | float hi_new = h[i] = momentum * hi + local_rate * g[i]; 12 | g[i] = (1+momentum) * hi_new - momentum * hi; 13 | } 14 | } 15 | template 16 | void nesterov_update_gpu(int N, Dtype* g, Dtype* h, Dtype momentum, 17 | Dtype local_rate) { 18 | NesterovUpdate // NOLINT_NEXT_LINE(whitespace/operators) 19 | <<>>( 20 | N, g, h, momentum, local_rate); 21 | CUDA_POST_KERNEL_CHECK; 22 | } 23 | template void nesterov_update_gpu(int, float*, float*, float, float); 24 | template void nesterov_update_gpu(int, double*, double*, double, 25 | double); 26 | 27 | } // namespace caffe 28 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/solvers/rmsprop_solver.cu: -------------------------------------------------------------------------------- 1 | #include "caffe/util/math_functions.hpp" 2 | 3 | 4 | namespace caffe { 5 | 6 | template 7 | __global__ void RMSPropUpdate(int N, Dtype* g, Dtype* h, 8 | Dtype rms_decay, Dtype delta, Dtype local_rate) { 9 | CUDA_KERNEL_LOOP(i, N) { 10 | float gi = g[i]; 11 | float hi = h[i] = rms_decay*h[i] + (1-rms_decay)*gi*gi; 12 | g[i] = local_rate * g[i] / (sqrt(hi) + delta); 13 | } 14 | } 15 | template 16 | void rmsprop_update_gpu(int N, Dtype* g, Dtype* h, Dtype rms_decay, 17 | Dtype delta, Dtype local_rate) { 18 | RMSPropUpdate // NOLINT_NEXT_LINE(whitespace/operators) 19 | <<>>( 20 | N, g, h, rms_decay, delta, local_rate); 21 | CUDA_POST_KERNEL_CHECK; 22 | } 23 | template void rmsprop_update_gpu(int, float*, float*, float, float, 24 | float); 25 | template void rmsprop_update_gpu(int, double*, double*, double, double, 26 | double); 27 | 28 | } // namespace caffe 29 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/solvers/sgd_solver.cu: -------------------------------------------------------------------------------- 1 | #include "caffe/util/math_functions.hpp" 2 | 3 | 4 | namespace caffe { 5 | 6 | template 7 | __global__ void SGDUpdate(int N, Dtype* g, Dtype* h, 8 | Dtype momentum, Dtype local_rate) { 9 | CUDA_KERNEL_LOOP(i, N) { 10 | g[i] = h[i] = momentum*h[i] + local_rate*g[i]; 11 | } 12 | } 13 | template 14 | void sgd_update_gpu(int N, Dtype* g, Dtype* h, Dtype momentum, 15 | Dtype local_rate) { 16 | SGDUpdate // NOLINT_NEXT_LINE(whitespace/operators) 17 | <<>>( 18 | N, g, h, momentum, local_rate); 19 | CUDA_POST_KERNEL_CHECK; 20 | } 21 | template void sgd_update_gpu(int, float*, float*, float, float); 22 | template void sgd_update_gpu(int, double*, double*, double, double); 23 | 24 | } // namespace caffe 25 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/test/CMakeLists.txt: -------------------------------------------------------------------------------- 1 | # The option allows to include in build only selected test files and exclude all others 2 | # Usage example: 3 | # cmake -DBUILD_only_tests="common,net,blob,im2col_kernel" 4 | set(BUILD_only_tests "" CACHE STRING "Blank or comma-separated list of test files to build without 'test_' prefix and extention") 5 | caffe_leave_only_selected_tests(test_srcs ${BUILD_only_tests}) 6 | caffe_leave_only_selected_tests(test_cuda ${BUILD_only_tests}) 7 | 8 | # For 'make runtest' target we don't need to embed test data paths to 9 | # source files, because test target is executed in source directory 10 | # That's why the lines below are commented. TODO: remove them 11 | 12 | # definition needed to include CMake generated files 13 | #add_definitions(-DCMAKE_BUILD) 14 | 15 | # generates test_data/sample_data_list.txt.gen.cmake 16 | #caffe_configure_testdatafile(test_data/sample_data_list.txt) 17 | 18 | set(the_target test.testbin) 19 | set(test_args --gtest_shuffle) 20 | 21 | if(HAVE_CUDA) 22 | caffe_cuda_compile(test_cuda_objs ${test_cuda}) 23 | list(APPEND test_srcs ${test_cuda_objs} ${test_cuda}) 24 | else() 25 | list(APPEND test_args --gtest_filter="-*GPU*") 26 | endif() 27 | 28 | # ---[ Adding test target 29 | add_executable(${the_target} EXCLUDE_FROM_ALL ${test_srcs}) 30 | target_link_libraries(${the_target} gtest ${Caffe_LINK}) 31 | caffe_default_properties(${the_target}) 32 | caffe_set_runtime_directory(${the_target} "${PROJECT_BINARY_DIR}/test") 33 | 34 | # ---[ Adding runtest 35 | add_custom_target(runtest COMMAND ${the_target} ${test_args} 36 | WORKING_DIRECTORY ${PROJECT_SOURCE_DIR}) 37 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/test/test_caffe_main.cpp: -------------------------------------------------------------------------------- 1 | // The main caffe test code. Your test cpp code should include this hpp 2 | // to allow a main function to be compiled into the binary. 3 | 4 | #include "caffe/caffe.hpp" 5 | #include "caffe/test/test_caffe_main.hpp" 6 | 7 | namespace caffe { 8 | #ifndef CPU_ONLY 9 | cudaDeviceProp CAFFE_TEST_CUDA_PROP; 10 | #endif 11 | } 12 | 13 | #ifndef CPU_ONLY 14 | using caffe::CAFFE_TEST_CUDA_PROP; 15 | #endif 16 | 17 | int main(int argc, char** argv) { 18 | ::testing::InitGoogleTest(&argc, argv); 19 | caffe::GlobalInit(&argc, &argv); 20 | #ifndef CPU_ONLY 21 | // Before starting testing, let's first print out a few cuda defice info. 22 | int device; 23 | cudaGetDeviceCount(&device); 24 | cout << "Cuda number of devices: " << device << endl; 25 | if (argc > 1) { 26 | // Use the given device 27 | device = atoi(argv[1]); 28 | cudaSetDevice(device); 29 | cout << "Setting to use device " << device << endl; 30 | } else if (CUDA_TEST_DEVICE >= 0) { 31 | // Use the device assigned in build configuration; but with a lower priority 32 | device = CUDA_TEST_DEVICE; 33 | } 34 | cudaGetDevice(&device); 35 | cout << "Current device id: " << device << endl; 36 | cudaGetDeviceProperties(&CAFFE_TEST_CUDA_PROP, device); 37 | cout << "Current device name: " << CAFFE_TEST_CUDA_PROP.name << endl; 38 | #endif 39 | // invoke the test. 40 | return RUN_ALL_TESTS(); 41 | } 42 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/test/test_data/sample_data.h5: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/src/caffe/test/test_data/sample_data.h5 -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/test/test_data/sample_data_2_gzip.h5: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/src/caffe/test/test_data/sample_data_2_gzip.h5 -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/test/test_data/sample_data_list.txt: -------------------------------------------------------------------------------- 1 | src/caffe/test/test_data/sample_data.h5 2 | src/caffe/test/test_data/sample_data_2_gzip.h5 3 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/test/test_data/solver_data.h5: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Xyuan13/MSRNet/ca18f9367691d82213006b1ffe0e207ac24f23ba/deeplab-caffe/src/caffe/test/test_data/solver_data.h5 -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/test/test_data/solver_data_list.txt: -------------------------------------------------------------------------------- 1 | src/caffe/test/test_data/solver_data.h5 2 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/test/test_internal_thread.cpp: -------------------------------------------------------------------------------- 1 | #include "glog/logging.h" 2 | #include "gtest/gtest.h" 3 | 4 | #include "caffe/internal_thread.hpp" 5 | #include "caffe/util/math_functions.hpp" 6 | 7 | #include "caffe/test/test_caffe_main.hpp" 8 | 9 | namespace caffe { 10 | 11 | 12 | class InternalThreadTest : public ::testing::Test {}; 13 | 14 | TEST_F(InternalThreadTest, TestStartAndExit) { 15 | InternalThread thread; 16 | EXPECT_FALSE(thread.is_started()); 17 | thread.StartInternalThread(); 18 | EXPECT_TRUE(thread.is_started()); 19 | thread.StopInternalThread(); 20 | EXPECT_FALSE(thread.is_started()); 21 | } 22 | 23 | class TestThreadA : public InternalThread { 24 | void InternalThreadEntry() { 25 | EXPECT_EQ(4244559767, caffe_rng_rand()); 26 | } 27 | }; 28 | 29 | class TestThreadB : public InternalThread { 30 | void InternalThreadEntry() { 31 | EXPECT_EQ(1726478280, caffe_rng_rand()); 32 | } 33 | }; 34 | 35 | TEST_F(InternalThreadTest, TestRandomSeed) { 36 | TestThreadA t1; 37 | Caffe::set_random_seed(9658361); 38 | t1.StartInternalThread(); 39 | t1.StopInternalThread(); 40 | 41 | TestThreadA t2; 42 | Caffe::set_random_seed(9658361); 43 | t2.StartInternalThread(); 44 | t2.StopInternalThread(); 45 | 46 | TestThreadB t3; 47 | Caffe::set_random_seed(3435563); 48 | t3.StartInternalThread(); 49 | t3.StopInternalThread(); 50 | } 51 | 52 | } // namespace caffe 53 | 54 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/test/test_protobuf.cpp: -------------------------------------------------------------------------------- 1 | // This is simply a script that tries serializing protocol buffer in text 2 | // format. Nothing special here and no actual code is being tested. 3 | #include 4 | 5 | #include "google/protobuf/text_format.h" 6 | #include "gtest/gtest.h" 7 | 8 | #include "caffe/proto/caffe.pb.h" 9 | 10 | #include "caffe/test/test_caffe_main.hpp" 11 | 12 | namespace caffe { 13 | 14 | class ProtoTest : public ::testing::Test {}; 15 | 16 | TEST_F(ProtoTest, TestSerialization) { 17 | LayerParameter param; 18 | param.set_name("test"); 19 | param.set_type("Test"); 20 | std::cout << "Printing in binary format." << std::endl; 21 | std::cout << param.SerializeAsString() << std::endl; 22 | std::cout << "Printing in text format." << std::endl; 23 | std::string str; 24 | google::protobuf::TextFormat::PrintToString(param, &str); 25 | std::cout << str << std::endl; 26 | EXPECT_TRUE(true); 27 | } 28 | 29 | } // namespace caffe 30 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/util/cudnn.cpp: -------------------------------------------------------------------------------- 1 | #ifdef USE_CUDNN 2 | #include "caffe/util/cudnn.hpp" 3 | 4 | namespace caffe { 5 | namespace cudnn { 6 | 7 | float dataType::oneval = 1.0; 8 | float dataType::zeroval = 0.0; 9 | const void* dataType::one = 10 | static_cast(&dataType::oneval); 11 | const void* dataType::zero = 12 | static_cast(&dataType::zeroval); 13 | 14 | double dataType::oneval = 1.0; 15 | double dataType::zeroval = 0.0; 16 | const void* dataType::one = 17 | static_cast(&dataType::oneval); 18 | const void* dataType::zero = 19 | static_cast(&dataType::zeroval); 20 | 21 | } // namespace cudnn 22 | } // namespace caffe 23 | #endif 24 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/util/db.cpp: -------------------------------------------------------------------------------- 1 | #include "caffe/util/db.hpp" 2 | #include "caffe/util/db_leveldb.hpp" 3 | #include "caffe/util/db_lmdb.hpp" 4 | 5 | #include 6 | 7 | namespace caffe { namespace db { 8 | 9 | DB* GetDB(DataParameter::DB backend) { 10 | switch (backend) { 11 | #ifdef USE_LEVELDB 12 | case DataParameter_DB_LEVELDB: 13 | return new LevelDB(); 14 | #endif // USE_LEVELDB 15 | #ifdef USE_LMDB 16 | case DataParameter_DB_LMDB: 17 | return new LMDB(); 18 | #endif // USE_LMDB 19 | default: 20 | LOG(FATAL) << "Unknown database backend"; 21 | return NULL; 22 | } 23 | } 24 | 25 | DB* GetDB(const string& backend) { 26 | #ifdef USE_LEVELDB 27 | if (backend == "leveldb") { 28 | return new LevelDB(); 29 | } 30 | #endif // USE_LEVELDB 31 | #ifdef USE_LMDB 32 | if (backend == "lmdb") { 33 | return new LMDB(); 34 | } 35 | #endif // USE_LMDB 36 | LOG(FATAL) << "Unknown database backend"; 37 | return NULL; 38 | } 39 | 40 | } // namespace db 41 | } // namespace caffe 42 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/util/db_leveldb.cpp: -------------------------------------------------------------------------------- 1 | #ifdef USE_LEVELDB 2 | #include "caffe/util/db_leveldb.hpp" 3 | 4 | #include 5 | 6 | namespace caffe { namespace db { 7 | 8 | void LevelDB::Open(const string& source, Mode mode) { 9 | leveldb::Options options; 10 | options.block_size = 65536; 11 | options.write_buffer_size = 268435456; 12 | options.max_open_files = 100; 13 | options.error_if_exists = mode == NEW; 14 | options.create_if_missing = mode != READ; 15 | leveldb::Status status = leveldb::DB::Open(options, source, &db_); 16 | CHECK(status.ok()) << "Failed to open leveldb " << source 17 | << std::endl << status.ToString(); 18 | LOG(INFO) << "Opened leveldb " << source; 19 | } 20 | 21 | } // namespace db 22 | } // namespace caffe 23 | #endif // USE_LEVELDB 24 | -------------------------------------------------------------------------------- /deeplab-caffe/src/caffe/util/densecrf_util.cpp: -------------------------------------------------------------------------------- 1 | #include 2 | 3 | #include "caffe/util/densecrf_util.hpp" 4 | 5 | float* allocate(size_t N) { 6 | float * r = NULL; 7 | if (N>0) { 8 | #ifdef SSE_DENSE_CRF 9 | r = (float*)_mm_malloc( N*sizeof(float)+16, 16 ); 10 | #else 11 | r = new float[N]; 12 | #endif 13 | } 14 | 15 | memset( r, 0, sizeof(float)*N); 16 | return r; 17 | } 18 | 19 | void deallocate(float*& ptr) { 20 | if (ptr) 21 | #ifdef SSE_DENSE_CRF 22 | _mm_free( ptr ); 23 | #else 24 | delete[] ptr; 25 | #endif 26 | ptr = NULL; 27 | } 28 | 29 | -------------------------------------------------------------------------------- /deeplab-caffe/src/gtest/CMakeLists.txt: -------------------------------------------------------------------------------- 1 | add_library(gtest STATIC EXCLUDE_FROM_ALL gtest.h gtest-all.cpp) 2 | caffe_default_properties(gtest) 3 | 4 | #add_library(gtest_main gtest_main.cc) 5 | #target_link_libraries(gtest_main gtest) 6 | -------------------------------------------------------------------------------- /deeplab-caffe/tools/CMakeLists.txt: -------------------------------------------------------------------------------- 1 | # Collect source files 2 | file(GLOB_RECURSE srcs ${CMAKE_CURRENT_SOURCE_DIR}/*.cpp) 3 | 4 | # Build each source file independently 5 | foreach(source ${srcs}) 6 | get_filename_component(name ${source} NAME_WE) 7 | 8 | # caffe target already exits 9 | if(name MATCHES "caffe") 10 | set(name ${name}.bin) 11 | endif() 12 | 13 | # target 14 | add_executable(${name} ${source}) 15 | target_link_libraries(${name} ${Caffe_LINK}) 16 | caffe_default_properties(${name}) 17 | 18 | # set back RUNTIME_OUTPUT_DIRECTORY 19 | caffe_set_runtime_directory(${name} "${PROJECT_BINARY_DIR}/tools") 20 | caffe_set_solution_folder(${name} tools) 21 | 22 | # restore output name without suffix 23 | if(name MATCHES "caffe.bin") 24 | set_target_properties(${name} PROPERTIES OUTPUT_NAME caffe) 25 | endif() 26 | 27 | # Install 28 | install(TARGETS ${name} DESTINATION bin) 29 | endforeach(source) 30 | -------------------------------------------------------------------------------- /deeplab-caffe/tools/device_query.cpp: -------------------------------------------------------------------------------- 1 | #include "caffe/common.hpp" 2 | 3 | int main(int argc, char** argv) { 4 | LOG(FATAL) << "Deprecated. Use caffe device_query " 5 | "[--device_id=0] instead."; 6 | return 0; 7 | } 8 | -------------------------------------------------------------------------------- /deeplab-caffe/tools/extra/launch_resize_and_crop_images.sh: -------------------------------------------------------------------------------- 1 | #!/bin/bash 2 | #### https://github.com/Yangqing/mincepie/wiki/Launch-Your-Mapreducer 3 | 4 | # If you encounter error that the address already in use, kill the process. 5 | # 11235 is the port of server process 6 | # https://github.com/Yangqing/mincepie/blob/master/mincepie/mince.py 7 | # sudo netstat -ap | grep 11235 8 | # The last column of the output is PID/Program name 9 | # kill -9 PID 10 | # Second solution: 11 | # nmap localhost 12 | # fuser -k 11235/tcp 13 | # Or just wait a few seconds. 14 | 15 | ## Launch your Mapreduce locally 16 | # num_clients: number of processes 17 | # image_lib: OpenCV or PIL, case insensitive. The default value is the faster OpenCV. 18 | # input: the file containing one image path relative to input_folder each line 19 | # input_folder: where are the original images 20 | # output_folder: where to save the resized and cropped images 21 | ./resize_and_crop_images.py --num_clients=8 --image_lib=opencv --input=/home/user/Datasets/ImageNet/ILSVRC2010/ILSVRC2010_images.txt --input_folder=/home/user/Datasets/ImageNet/ILSVRC2010/ILSVRC2010_images_train/ --output_folder=/home/user/Datasets/ImageNet/ILSVRC2010/ILSVRC2010_images_train_resized/ 22 | 23 | ## Launch your Mapreduce with MPI 24 | # mpirun -n 8 --launch=mpi resize_and_crop_images.py --image_lib=opencv --input=/home/user/Datasets/ImageNet/ILSVRC2010/ILSVRC2010_images.txt --input_folder=/home/user/Datasets/ImageNet/ILSVRC2010/ILSVRC2010_images_train/ --output_folder=/home/user/Datasets/ImageNet/ILSVRC2010/ILSVRC2010_images_train_resized/ 25 | -------------------------------------------------------------------------------- /deeplab-caffe/tools/finetune_net.cpp: -------------------------------------------------------------------------------- 1 | #include "caffe/caffe.hpp" 2 | 3 | int main(int argc, char** argv) { 4 | LOG(FATAL) << "Deprecated. Use caffe train --solver=... " 5 | "[--weights=...] instead."; 6 | return 0; 7 | } 8 | -------------------------------------------------------------------------------- /deeplab-caffe/tools/net_speed_benchmark.cpp: -------------------------------------------------------------------------------- 1 | #include "caffe/caffe.hpp" 2 | 3 | int main(int argc, char** argv) { 4 | LOG(FATAL) << "Deprecated. Use caffe time --model=... " 5 | "[--iterations=50] [--gpu] [--device_id=0]"; 6 | return 0; 7 | } 8 | -------------------------------------------------------------------------------- /deeplab-caffe/tools/test_net.cpp: -------------------------------------------------------------------------------- 1 | #include "caffe/caffe.hpp" 2 | 3 | int main(int argc, char** argv) { 4 | LOG(FATAL) << "Deprecated. Use caffe test --model=... " 5 | "--weights=... instead."; 6 | return 0; 7 | } 8 | -------------------------------------------------------------------------------- /deeplab-caffe/tools/test_read_label.cpp: -------------------------------------------------------------------------------- 1 | #include // NOLINT(readability/streams) 2 | #include // NOLINT(readability/streams) 3 | #include 4 | #include 5 | #include 6 | #include 7 | 8 | #include 9 | #include 10 | #include 11 | #include 12 | 13 | #include "caffe/data_transformer.hpp" 14 | #include "caffe/layers/base_data_layer.hpp" 15 | #include "caffe/layers/image_seg_data_layer.hpp" 16 | #include "caffe/util/benchmark.hpp" 17 | #include "caffe/util/io.hpp" 18 | #include "caffe/util/math_functions.hpp" 19 | #include "caffe/util/rng.hpp" 20 | 21 | 22 | int main() { 23 | 24 | std::string root_folder="/home/phoenix/Dataset/MSRA-B"; 25 | std::string img_id="/annotation/9_58060.png"; 26 | int new_height=513; 27 | int new_width=513; 28 | bool is_color=false; 29 | // Read an image, and use it to initialize the top blob. 30 | cv::Mat cv_img = caffe::ReadImageToCVMat(root_folder + img_id, 31 | new_height, new_width, is_color); 32 | CHECK(cv_img.data) << "Could not load " << img_id; 33 | 34 | std::cout << cv_img.rows << std::endl; 35 | std::cout << cv_img.cols << std::endl; 36 | std::cout << cv_img.channels(); 37 | //cv::imshow("label", cv_img); 38 | //cv::waitKey(0); 39 | /* for (int h = 0; h < new_height; h++) { 40 | for (int w = 0; w < new_width; w++) 41 | std::cout<< int(cv_img.at(h,w))<<" "; 42 | std::cout << std::endl; 43 | } 44 | */ return 0; 45 | } 46 | 47 | -------------------------------------------------------------------------------- /deeplab-caffe/tools/train_net.cpp: -------------------------------------------------------------------------------- 1 | #include "caffe/caffe.hpp" 2 | 3 | int main(int argc, char** argv) { 4 | LOG(FATAL) << "Deprecated. Use caffe train --solver=... " 5 | "[--snapshot=...] instead."; 6 | return 0; 7 | } 8 | -------------------------------------------------------------------------------- /models_prototxts/get_msrnet-vgg_model.sh: -------------------------------------------------------------------------------- 1 | cd ./models_prototxts 2 | echo 'Download model and prototxt ------------------' 3 | wget http://www.sysu-hcp.net/wp-content/uploads/2017/09/MSRNet-VGG_models_prototxts.zip 4 | echo 'Successfully download the models and prototxts ------------------' 5 | cd ../ 6 | --------------------------------------------------------------------------------